CN108985163A - Confined Space Safety Detection Method Based on Unmanned Aerial Vehicle - Google Patents

Abstract

The application relates to a limited space safety detection method based on an unmanned aerial vehicle, which adopts an information acquisition device of the unmanned aerial vehicle for safety monitoring information acquisition of an area needing safety monitoring detection in a limited space, and outputs an identification result after the acquired safety monitoring information is subjected to safety monitoring information identification operation on a safety monitoring information identification module by using a safety monitoring information identification method. And the safety monitoring work of the limited space is safer and more intelligent.

Description

Confined Space Safety Detection Method Based on Unmanned Aerial Vehicle

【Technical field】

The present application relates to the field of safety detection, and in particular to a method for safe detection in a restricted space based on an unmanned aerial vehicle.

【Background technique】

In the energy industry, such as coal mining, it is still necessary to rely on professionals to enter in advance for inspection to ensure safety. This is very dangerous for inspectors, and the uncertainty of the inspection environment is very large. Another example is petrochemical safety. , petrochemical oil storage tanks (used to store gasoline, diesel and natural gas, etc.) The gas detection device is tied to a long object such as a bamboo pole and then stretched into the oil storage tank, and then taken out to check the gas content indicators inside. If the indicators are qualified, relevant personnel will be further allowed to enter the oil storage tank to check the health of the oil storage tank. Whether the tank has cracks, rust, etc. And the method that this detection method puts the gas detection device in and then takes it out is defective, because when the gas detection device is taken out, the gas content index will change to some extent than when it is inside, and the accuracy of the data is insufficient and The detection area is also very limited. In addition, there is a great danger in allowing personnel to work in such an uncertain environment.

The above-mentioned inspected environment can be called a restricted space. The restricted space also includes various buildings or equipment that are difficult for people to reach or have a high risk index, such as various equipment in a factory or the interior of a building or the exterior that is difficult to reach. (such as factory furnaces, tower kettles, tanks, warehouses, pools, tank cars, pipes, flues, etc.) and urban tunnels, sewers, ditches, pits, wells, pools, culverts, valve rooms, sewage treatment facilities, etc. Semi-enclosed facilities and places (such as cabins, underground hidden projects, airtight containers, facilities that have not been used for a long time or places with poor ventilation, etc.), as well as wells and cellars for storing sweet potatoes, potatoes, and various vegetables in rural areas. Poorly ventilated mine shafts should also be considered confined spaces.

Based on this, there is a need for a new method that can replace or assist humans to detect confined spaces.

【Content of invention】

In order to solve the above technical problems, the purpose of this application is to provide a method for safe detection of restricted spaces based on unmanned aerial vehicles, and use unmanned aerial vehicles to replace personnel to detect the situation in restricted spaces, which greatly reduces the unsafe situation of personnel entering restricted spaces. The detection is also very convenient, and the timeliness and stability of the judgment results are enhanced. According to the degree of defects detected in the facilities in the confined space, it is possible to choose whether to immediately arrange personnel for maintenance or continue to use the facilities, because the degree of individual defects does not necessarily affect the work of the facilities, and it is better to wait until the defects reach a certain level before performing maintenance. It is conducive to improving the effective utilization of resources and making the safety monitoring work in confined spaces safer and smarter.

The application is achieved through the following technical solutions:

Based on the UAV-based restricted space safety detection method, the restricted space safety detection UAV information collection device is used to collect safety monitoring information for areas that require safety monitoring and detection in the restricted space, and the collected safety monitoring information is used In the safety monitoring information recognition method, the safety monitoring information recognition operation is performed on the safety monitoring information recognition module, and the recognition result is output.

In the UAV-based restricted space safety detection method described above, the information collection device is a camera, the safety monitoring information is image information, the safety monitoring information recognition module is an image recognition module, and the safety monitoring information The recognition method is an image recognition method, the image information collected by the camera is used on the image recognition module to output the recognition result after the image recognition method is used to recognize and operate the image information, and the image recognition method includes the following steps:

S21, classify the detection scenes in the confined space, and make a training data set by matching the identification images of the classified defects for each type of detection scene, the detection scenes are classified into gas pipeline burners, support storage tanks, Inside industrial chimneys, large oil tank ships, large pipelines or coal-fired power plant boilers, the degree of defect is severe corrosion, moderate corrosion or mild corrosion, and the identification result is severe corrosion, moderate corrosion or mild corrosion ;

S22, inputting the identification image that has classified the detection scene and defect degree in the training data set in step S21, and performing image sharpening and edge enhancement processing;

S23, using the training data set formed by the identification image processed in step S22 to train the convolutional neural network;

S24, input the image information collected in step S1 into the convolutional neural network trained in step S22, and output the recognition result.

As described above, based on the UAV-based limited space safety detection method, the training of the convolutional neural network described in step S23 includes the following steps:

S231, performing the first convolution processing on the identification images in the training data set processed in step S22, forming the first layer of the convolutional neural network;

S232: Perform the first pooling process on the first layer of the convolutional neural network, and perform local response normalization processing to form the second layer of the convolutional neural network;

S233. Perform a second convolution process on the second layer of the convolutional neural network to form the third layer of the convolutional neural network;

S234, performing a second pooling process on the third layer of the convolutional neural network to form the fourth layer of the convolutional neural network;

S235. Obtain the fifth layer of the convolutional neural network in a fully connected manner for the fourth layer of the convolutional neural network;

S236, the fifth layer of the convolutional neural network is fully connected to obtain an output layer: the output layer is divided into 6 independent parts, and each part is additionally connected to an independent loss function;

S237. Minimize the value of the loss function by using the identified identification images that have been classified in S21, and complete the training of the convolutional neural network.

In the UAV-based restricted space safety detection method described above, the size of the identification image processed in step S22 is unified to 32×32×1, the filter size of the convolutional layer is 5×5, and the depth is 6 , do not use all 0 supplements, the step size is 1, and the output size of the first layer of the convolutional neural network is 32-5+1=28, and the depth is 6, that is, the node matrix of the second layer of the convolutional neural network There are 28×28×6=4704 nodes, and each node is connected to 5×5=25 current layer nodes, that is, the first layer of the convolutional neural network has a total of 4704×(25+1)=122304 connections;

The size of the filter used in the second layer of the convolutional neural network is 2×2, and the step size is 2, and the size of the output matrix of the second layer of the convolutional neural network is 14×14×6;

The output matrix of the second layer of the convolutional neural network is the input of the third layer of the convolutional neural network. The size of the filter used is 5×5, the depth is 16, and all 0s are not used to supplement, the step size is 1, and the volume The output size of the third layer of the convolutional neural network is 14-5+1=10, and the depth is 16, that is, the size of the output matrix of the third layer of the convolutional neural network is 10×10×16, and the parameters of the third layer of the convolutional neural network are 5 ×5×6×16+16=2416, there are 10×10×16×(5×5+1)=41600 connections;

The size of the input matrix of the fourth layer of the convolutional neural network is 10×10×16, the size of the filter used is 2×2, and the step size is 2, and the size of the output matrix of the fourth layer of the convolutional neural network is 5 ×5×16;

The output matrix of the fourth layer of the convolutional neural network is the input of the fifth layer of the convolutional neural network, and it is straightened into a vector with a length of 5×5×16, that is, a three-dimensional matrix is pulled into a one-dimensional space Represented in the form of a vector, enter the fully connected layer for training, the number of output nodes of the fifth layer of the convolutional neural network is 120, that is, there are 5×5×16×120+120=48120 parameters in total.

In the above-mentioned UAV-based restricted space safety detection method, the information collection device also includes a gas sensor for collecting gas information in areas that require safety monitoring and detection in the restricted space, and the safety monitoring information is gas information, the safety monitoring information identification module is a gas identification module, the safety monitoring information identification method is a gas identification method, and the gas information collected by the gas sensor uses the gas identification method on the gas identification module to analyze the gas information Output the recognition result after the recognition operation.

As described above, the UAV-based restricted space safety detection method, the UAV includes a UAV body and an anti-collision frame arranged outside the UAV body; the UAV The machine body includes a frame and a cloud platform installed above the frame, the frame is provided with a down-mounted rotor, the camera is arranged on the platform, and one side of the camera is provided with the The illuminating light that the camera moves; the anti-collision frame includes a first hemispherical frame, a second hemispherical frame and a connecting rod, and the cutting edges of the first hemispherical frame and the second hemispherical frame are connected to each other by the connecting rod to form a The anti-collision frame.

According to the UAV-based restricted space safety detection method, a plurality of connecting rods are evenly distributed along the edge of the cutting plane of the first hemispherical frame and the second hemispherical frame to form a connection between the first hemispherical frame and the second hemispherical frame. The cylindrical frame of the second hemispherical frame.

According to the UAV-based method for safe detection of confined spaces as described above, the frame includes a base, and the two sides of the base are symmetrically provided with installation frames, and the installation frames include eight The first rod and the second rod extending out from one side of the base in the shape of a font, the third rod connected with the extended end of the first rod, and the fourth rod connected with the extended end of the second rod , the free ends of the third rod and the fourth rod intersect to form a corner connection part connected with the anti-collision frame, and the lower-mounted rotor is arranged at the connection between the first rod and the third rod below the part and below the connecting part of the second rod and the fourth rod;

The first rod, the second rod, the third rod and the fourth rod are hinged end to end to form a rhombus-shaped installation frame, and there is a gap between the inner corners of the first rod and the second rod. The stop protrusion provided on the base is used to pull the first rod and the second rod in opposite directions so that the first rod and the second rod are close to the stop A protruding first return spring, a second return spring is provided between the inner corners of the third rod and the fourth rod.

As mentioned above, the UAV-based limited space safety detection method, the first hemispherical frame includes:

There are a plurality of frame strips, one end of the plurality of frame strips is folded, and the other end is opened to form a hemispherical frame shape, and one end of the frame strip is provided with a protrusion protruding outward from the ball top of the first hemispherical frame, The open end of the frame bar is provided with a recess that is recessed toward the center of the cut surface of the first hemispherical frame;

Gathering ring, which is arranged at one end of the frame strip, the said gathering ring comprises an annular hole, an annular gathering wall surrounding said annular hole, annular grooves evenly distributed along said annular gathering wall, and an annular groove set in said annular The outer ring thread of the outer wall of the wall, the frame bars are embedded in the annular grooves and the locking protrusions are folded in the annular holes so that one end of a plurality of the frame bars is in a folded state;

A screw-in ring, which is provided with a fixing hole fixed to the frame, an annular screw-in wall surrounding the fixing hole, and an inner ring thread provided on the inner wall of the ring-shaped screw-in wall to match the outer ring thread, The gathering ring is screwed into the screwing ring after gathering the frame bar;

The opening ring is evenly provided with clamping holes equal in number to the frame strips and lock holes equal in number to the connecting rods along the periphery, and the opened end of the frame strips is clamped into the clamping holes;

The connecting ring is uniformly provided with through holes equal to the number of connecting rods along the circumference;

A connecting screw, the connecting screw passes through the locking hole and the through hole to lock the expansion ring and the connecting ring on the connecting rod.

In the above-mentioned UAV-based restricted space safety detection method, the restricted space safety detection UAV is also provided with a GPS for transmitting the position information of the restricted space safety detection UAV to the control terminal in real time A module and a marking gun for marking a fault location in a confined space, the marking gun is an automatic spray gun capable of spraying a colored solution.

Compared with the prior art, the present application has the following advantages:

1. UAVs are used instead of personnel to detect the situation in confined spaces, which greatly reduces the insecurity of personnel entering confined spaces, and the detection is also very convenient, and the timeliness and stability of the judgment results are enhanced. According to the degree of defects detected in the facilities in the confined space, it is possible to choose whether to immediately arrange personnel for maintenance or continue to use the facilities, because the degree of individual defects does not necessarily affect the work of the facilities, and it is better to wait until the defects reach a certain level before performing maintenance. It is conducive to improving the effective utilization of resources and making the safety monitoring work in confined spaces safer and smarter.

2. The lightweight anti-collision frame can rebound or rotate when it collides with an object, which can effectively prevent the UAV, especially the rotor, from being damaged. Flying in an unfamiliar and complex environment greatly reduces the incidence of "flying planes", ensures that drones can normally collect information in confined spaces and greatly reduces the difficulty of controlling drones.

3. The upper part of the lower-mounted rotor is connected to the output shaft of the coaxial brushless DC motor. Although the stability of the lower-mounted rotor has decreased, the maneuverability of the UAV is stronger, which is more conducive to its use in complex confined spaces. Avoid obstacles in time, improve the safety of the UAV, and make it easier for the camera to be installed above the UAV to capture a wider upper image.

4. A cylindrical frame is set between the first hemispherical frame and the second hemispherical frame, so that when the anti-collision frame falls to the ground, on the one hand, it can still offset part of the impact force by rolling, on the other hand, the round The cylindrical frame can roll in line contact with the ground to prevent the long-term unstable point contact rolling between the circular anti-collision frame and the ground, which will cause multi-directional irregular flips on the drone body and increase the chance of damage.

【Description of drawings】

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those skilled in the art, other drawings can also be obtained based on these drawings without creative effort.

Fig. 1 is the structural representation 1 of embodiment 1 of the present application;

Fig. 2 is the schematic structural diagram 2 of embodiment 1 of the present application;

Fig. 3 is an enlarged view of part A of Fig. 2;

Fig. 4 is an enlarged view of part B of Fig. 2;

FIG. 5 is a schematic diagram of the decomposition structure of Embodiment 1 of the present application;

Fig. 6 is a schematic diagram of an exploded structure of the gathering ring and the screwing ring described in Embodiment 1 of the present application;

FIG. 7 is a schematic structural view of the rack described in Embodiment 1 of the present application;

Figure 8 is a schematic structural view of the marking gun described in Embodiment 1 of the present application;

Fig. 9 is a schematic diagram of the outlet of the nozzle described in Embodiment 1 of the present application;

Fig. 10 is a schematic structural diagram of the rack described in Embodiment 2 of the present application;

FIG. 11 is an image information diagram captured in a confined space in Embodiment 1 of the present application;

FIG. 12 is an image information diagram of the image information in FIG. 11 after being processed in step S22 described in Embodiment 1;

FIG. 13 is a schematic diagram of image recognition described in Embodiment 1 of the present application;

FIG. 14 is a schematic diagram of the principle of convolution processing described in Embodiment 1 of the present application;

FIG. 15 is a schematic diagram of the principle of the pooling process described in Embodiment 1 of the present application.

【Detailed ways】

In order to make the technical problems, technical solutions and beneficial effects solved by the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, and are not intended to limit the present application.

As shown in Figures 1 to 10, the embodiment of the present application proposes a restricted space safety detection drone, including a drone body 1 and an anti-collision frame 2 arranged outside the drone body 1; Machine body 1 comprises frame 11 and the cloud platform 12 that is installed on described frame top, and described frame 11 is provided with down-mounted rotor 110, and described cloud platform 12 is provided with camera 121, moves with described camera 121 The illuminating lamp 122 and the GPS module used to transmit the position information of the drone to the control terminal in real time; The frame 21 and the cut edge of the second hemispherical frame 22 are connected to each other by the connecting rod 23 to form the anti-collision frame 2 . The above-mentioned restricted space safety detection UAV can replace human beings to carry out information collection, safety inspection and other work in the restricted space.

An obstacle avoidance module can also be set in the drone body, such as an infrared probe or an ultrasonic probe, which is conducive to the safe detection of a confined space. The drone avoids obstacles in a complex confined space. The control level requirements are still relatively high, and the above-mentioned anti-collision framework can greatly reduce the difficulty of controlling the UAV for safe detection in confined spaces, and protect UAVs for safe detection in confined spaces to a greater extent. Specifically, the overall maximum diameter of the anti-collision frame 2 is less than 500mm, and its material is carbon fiber. The lightweight anti-collision frame can rebound or rotate when it collides with an object, which can effectively prevent the safe detection of drones in confined spaces, especially The rotor is not damaged, and it can even fall freely from a high altitude without damage, which is more conducive to the safe detection of confined spaces. UAVs can fly in unfamiliar and complex environments, greatly reducing the incidence of "bombing" and ensuring Confined space safety detection UAV can normally collect information in confined space.

In a confined space, the conditions are often complex, dark and airtight, and it is more likely to cause the UAV to collide with obstacles and fail to fly. The light that moves with the camera can illuminate the front, which is more conducive to the control terminal according to the The image transmitted by the camera controls the flight of the drone for safe detection in confined spaces, and at the same time helps the safe detection of drones in confined spaces to collect image information more clearly.

More specifically, high-intensity LED lights are selected for the illumination light and installed on the side of the camera. In addition, the drone body is equipped with a large-capacity battery, preferably a lithium battery as an energy supply. Furthermore, Feiyu mini3Dpro can be selected as the pan/tilt, which can be used as a supporting device for installing and fixing the camera, and can even adjust the horizontal and pitch angles of the camera. 3DR wireless data transmission module, WIFI wireless data transmission module or 4G network transmission module can be selected for data transmission of UAVs for safe detection in confined spaces. The control terminal includes a remote control and a monitor. The remote control can control the flight status of the drone for safe detection in a restricted space, and the monitor can instantly obtain the image information, location information, and other environments of the drone for safe detection in a confined space. information.

The upper part of the undermounted rotor is connected to the output shaft of the coaxial brushless DC motor. Although the stability of the undermounted rotor has decreased, the maneuverability of the UAV for safe detection in confined spaces is stronger, which is more conducive to its ability to operate in complex environments. Avoid obstacles in a confined space in time, improve the safety of the restricted space safety detection UAV, and make it easier for the camera to be installed above the restricted space safety detection UAV to capture a wider upper image.

Preferably, a plurality of connecting rods 23 are evenly distributed along the edge of the cutting surface of the first hemispherical frame 21 and the second hemispherical frame 22 to form a connecting rod between the first hemispherical frame 21 and the second hemispherical frame 22 Cylindrical frame. Specifically, there are four connecting rods 23, and a cylindrical frame is set between the first hemispherical frame and the second hemispherical frame, so that when the anti-collision frame falls to the ground, on the one hand, it still Part of the impact force can be offset by rolling. On the one hand, the cylindrical frame can roll in line contact with the ground to prevent the long-term unstable point contact rolling between the circular anti-collision frame and the ground, causing multi-directional damage to the drone body. And irregular flips increase the chance of damage.

In the first embodiment, the frame 11 includes a base 111, and the two sides of the base 111 are symmetrically provided with installation frames 112. The first rod 113 and the second rod 114 protruding from one side of the base 111, the third rod 115 connected with the extended end of the first rod 113 and the third rod 115 connected with the extended end of the second rod 114 The fourth rod 116, the free ends of the third rod 115 and the fourth rod 116 intersect to form the corner connection part 117 connected with the anti-collision frame 2, and the lower-mounted rotor 110 is arranged on the first Below the connecting portion of the rod 113 and the third rod 115 and below the connecting portion of the second rod 114 and the fourth rod 116 . Specifically, the first rod 113, the second rod 114, the third rod 115 and the fourth rod 116 are provided with a plurality of process holes, which can effectively reduce the weight of the frame, and the frame-shaped frame is on both sides of the base. Four installation points for installing the rotor are symmetrically formed, the structure is simple and lightweight, and it is beneficial to the reliable and stable installation of the rotor.

In this embodiment 1, the first hemispherical frame 21 includes: a frame bar 211, which is provided with a plurality of frame bars 211, one end of which is gathered, and the other end is opened to form a hemispherical frame shape, and the frame bars 211 are gathered One end is provided with a card protrusion 2111 protruding outward from the top of the first hemispherical frame 21, and one end of the frame bar 211 opened is provided with a card concave 2112 recessed toward the center of the cut surface of the first hemispherical frame 21; Ring 212, which is arranged at one end of the frame bar 211 to be folded, and the folded ring 212 includes an annular hole 2121, an annular folded wall 2122 surrounding the annular hole 2121, and annular grooves evenly distributed along the annular folded wall 2122 2123 and the outer ring thread 2124 provided on the outer wall of the annular wall 1142, the frame bar 211 is inserted into the annular groove 2123 and the protrusion 2111 is folded in the annular hole 2121 so that a plurality of the One end 5 of the frame bar 211 forms a folded state; the screw ring 213 is provided with a fixing hole 2131 fixed with the frame 11, an annular screw wall 2132 surrounding the fixed hole 2131 and an annular screw wall 2132 The inner ring thread 2133 matched with the outer ring thread 2124 on the inner wall, the gathering ring 212 is screwed into the screwing ring 213 after gathering the frame bar 211; The clamping holes 2141 with the same number of frame bars 211 and the lock holes 2142 with the same number as the connecting rods 23, the opened end of the frame bar 211 is inserted into the clamping holes 2141; the connecting ring 215, along the circumference The through holes equal to the number of the connecting rods 23 are evenly provided on the upper part; the connecting screws pass through the locking holes 2142 and the through holes to lock the expansion ring 214 and the connecting ring 215 in place. on the connecting rod 23.

The structure of the second hemispherical frame 22 is symmetrical to that of the first hemispherical frame 21 . The spherical structure makes the confined space safety detection UAV more uniform in force when it is collided, and the protection of the confined space safety detection UAV is more comprehensive, and the structure is firm and easy to disassemble.

The UAV for safety detection in a confined space also includes a marking gun 3 for marking a fault location in a confined space. When the restricted space safety detection drone finds the faulty part, it can use a marking gun to mark the faulty part, such as marking an X-shaped or triangular pattern, or even a pattern with a fluorescent effect to make the mark more prominent. Of course, the mark can be used At or near the fault location, combined with the positioning of the GPS module, personnel can find the fault location more intuitively, accurately and quickly when performing maintenance on facilities in confined spaces.

Preferably, the marking gun 3 is an automatic spray gun capable of spraying colored solution. The colored solution can be selected as a red solution, of course, fluorescent powder can be added to the solution to make the marking pattern more vivid. Specifically, the automatic spray gun includes: a nozzle 31, which is provided with an X-shaped outlet so that the sprayed colored solution is distributed in an X shape; a liquid spray channel 32, whose outlet is communicated with the nozzle 31; a solution bottle 33, It is used to hold the colored solution, which is provided with a liquid suction channel 34 for the liquid to flow to the bottle mouth, and the solution bottle 33 is provided with a first one-way valve 30 for air to enter the bottle; the piston chamber 35, It communicates with the liquid spray channel 32 and the liquid suction channel 34, and a control band solution is provided between the piston chamber 35 and the entrance of the liquid spray channel 32 to flow from the piston chamber 35 to the nozzle 31. A second one-way valve 36 that flows in the same direction, and a third one-way valve that controls the flow of the colored solution from the solution bottle 33 to the direction of the piston cavity 35 is provided between the piston cavity 35 and the liquid suction channel 34 37, the piston chamber 35 is provided with a piston 38 that can reciprocate in the piston chamber 35, and the middle part of the piston 38 is provided with a piston threaded hole; the driving motor 39 is provided with a motor shaft 391 with external threads , the motor shaft is screwed into the threaded hole of the piston to control the reciprocating motion of the piston 38 .

When working, the driving motor rotates forward, the driving piston shrinks back in the piston chamber to absorb the solution in the solution bottle into the piston chamber, then the driving motor reverses, and the driving piston moves forward quickly in the piston chamber to suck the solution in the piston chamber The solution is sprayed out through the nozzle.

The cloud platform 12 is also provided with a gas sensor, and multiple sensors can be integrated inside the gas sensor, including such as carbon monoxide gas sensor, gas gas sensor, Freon gas sensor, etc., which can measure the type, concentration and composition of gas , and convert it into an electrical signal. On the one hand, it detects safety issues such as gas pollution and gas leakage in confined spaces, and on the other hand, it detects the degree of gas hazard in confined spaces in advance to improve the safety of subsequent personnel entering confined spaces for maintenance.

It should be noted that the drone body is also provided with a main control board, and the electrical components such as the camera, GPS module, illumination lamp, drive motor, pan/tilt, gas sensor, and motor for controlling the rotation of the rotor are all connected with the main control board. The control board is electrically connected to realize control, and realize actions such as switching and turning.

This embodiment also provides a UAV-based restricted space safety detection method, including using an information collection device of a UAV for UAV safety detection in a restricted space to collect safety monitoring information for areas that require safety monitoring and detection in a restricted space, and collecting all The collected safety monitoring information uses the safety monitoring information identification method to carry out the safety monitoring information identification operation on the safety monitoring information identification module, and then outputs the identification result. Using unmanned aerial vehicles instead of personnel to detect the situation in confined spaces greatly reduces the insecurity of personnel entering confined spaces, the detection is also very convenient, and the timeliness and stability of the judgment results are enhanced. According to the degree of defects detected in the facilities in the confined space, it is possible to choose whether to immediately arrange personnel for maintenance or continue to use the facilities, because the degree of individual defects does not necessarily affect the work of the facilities, and it is better to wait until the defects reach a certain level before performing maintenance. It is conducive to improving the effective utilization of resources and making the safety monitoring work in confined spaces safer and smarter.

Correspondingly, the information collection device is a camera, the safety monitoring information is image information, the safety monitoring information recognition module is an image recognition module, the safety monitoring information recognition method is an image recognition method, and the information collected by the camera is The image information is recognized and calculated by the image recognition method on the image recognition module, and the recognition result is output. Of course, the restricted space safety detection drone can automatically cruise according to the set route, or it can be manually controlled in real time. Detecting drones to carry out camera activities in a restricted space, as shown in FIG. 11 , can be image information 600 captured in a restricted space. Figure 13 is a schematic diagram of image recognition, in which the identified corrosion locations are inside the rectangular frame. The image recognition method comprises the following steps:

S21, classify the detection scenes in the limited space, and match the identification images with the classified defect degree for each type of detection scene to make a training data set. It should be noted that the defects in the defect degree may include damage , cracks, corrosion, etc., such as pipeline leakage, oil storage tank damage, coal mine collapse, etc., and the safety monitoring information identification module can detect the above defects and output identification results corresponding to the degree, such as different degrees of leakage, different degrees of Damage, etc. In this embodiment, the detection scene is classified into gas pipeline burners, support storage tanks, inside of industrial chimneys, inside of large oil tank ships, large pipelines or inside of coal-fired power plant boilers, and the degree of defect is severe corrosion , moderate corrosion or mild corrosion, the identification result is severe corrosion, moderate corrosion or mild corrosion;

S22, input the identification image that has classified the detection scene and defect degree in the training data set in step S21, and perform image sharpening and edge enhancement processing, wherein, Fig. 12 is the image information 600 captured in the restricted space after step S22 processed image information 700;

S23, using the training data set formed by the identification image processed in step S22 to train the convolutional neural network;

S24. Input the image information collected in step S1 into the convolutional neural network trained in step S22, and output the recognition result. In this embodiment, the output value of the recognition result is severe corrosion, moderate corrosion or mild corrosion. corrosion.

Wherein, described in step S23, training the convolutional neural network includes the following steps:

S231, sequentially perform the first convolution processing on the identification images in the training data set processed in step S22, which is the first layer of the convolutional neural network. Preferably, the size of the identification images processed in step S22 is unified to 32 ×32×1, the filter size of the convolutional layer is 5×5, the depth is 6, all 0s are not used, the step size is 1, and the output size of the first layer of the convolutional neural network is 32-5+ 1=28, the depth is 6, that is, the second layer node matrix of the convolutional neural network has 28×28×6=4704 nodes, and each node is connected to 5×5=25 current layer nodes, that is, the convolutional neural network There are a total of 4704×(25+1)=122304 connections in the first layer of the network;

S232: Perform pooling processing on the first layer of the convolutional neural network for the first time, and local response normalization processing, which is the second layer of the convolutional neural network. Preferably, the second layer of the convolutional neural network adopts The filter size is 2×2, the step size is 2, and the output matrix size of the second layer of the convolutional neural network is 14×14×6;

S233. Perform a second convolution process on the second layer of the convolutional neural network, which is the third layer of the convolutional neural network. Preferably, the output matrix of the second layer of the convolutional neural network is the convolutional neural network For the input of the third layer, the filter size used is 5×5, the depth is 16, all zeros are not used, and the step size is 1. The output size of the third layer of the convolutional neural network is 14-5+1=10, The depth is 16, that is, the size of the output matrix of the third layer of the convolutional neural network is 10×10×16, and the parameters of the third layer of the convolutional neural network are 5×5×6×16+16=2416, and there are 10×10×16 ×(5×5+1)=41600 connections;

S234. Perform a second pooling process on the third layer of the convolutional neural network, which is the fourth layer of the convolutional neural network. The optimal degree, the size of the input matrix of the fourth layer of the convolutional neural network is 10×10× 16. The size of the filter used is 2×2, the step size is 2, and the size of the output matrix of the fourth layer of the convolutional neural network is 5×5×16;

S235. Obtain the fifth layer of the convolutional neural network in a fully connected manner for the fourth layer of the convolutional neural network. Preferably, the output matrix of the fourth layer of the convolutional neural network is the fifth layer of the convolutional neural network. The input of the layer is straightened into a vector with a length of 5×5×16, that is, a three-dimensional matrix is pulled to a one-dimensional space and expressed in the form of a vector, and it enters the fully connected layer for training. The fifth convolutional neural network The number of output nodes of the layer is 120, that is, there are 5×5×16×120+120=48120 parameters in total;

S236, the fifth layer of the convolutional neural network is fully connected to obtain an output layer: the output layer is divided into 6 independent parts, and each part is additionally connected to an independent loss function;

S237. Minimize the value of the loss function by using the classified identification images in S21 to complete the training of the convolutional neural network. This step is to further correct and optimize the convolutional neural network through the classified identification images.

Wherein, the sharpening processing method can be high-pass filtering and spatial differentiation method, and the edge enhancement processing can use the spatial domain method and the frequency domain method to highlight the edges where the brightness values of adjacent pixels of the image differ greatly, so as to efficiently Extract the features of image information accurately and enhance the accuracy of image recognition. The principle of the convolution processing is to bring the input data 100 into the convolution kernel algorithm, and convert it into a weight matrix 300 through the convolution kernel 200. It should be noted that the convolution kernel is the convolution layer mentioned above. filter, this matrix is combined with the image, resulting in a convolved output. The pooling process is used to reduce the number of training parameters of the convolutional neural network and the spatial size of the image. The pooling is completed independently on each depth dimension, so the convolutional layer is reduced under the premise that the depth of the image remains unchanged. The calculation amount, the principle is to select the maximum value of the non-overlapping regions in the image data to be pooled 400 to form the new image data 500 .

The above-mentioned UAV-based restricted space safety detection method uses UAVs instead of personnel to detect the situation in the restricted space, which greatly reduces the insecurity of personnel entering the restricted space, and the detection is also very convenient and intelligent, and enhances the Judge the timeliness and stability of the results. Moreover, in order to improve the effective utilization of resources, according to the degree of damage or defect of the detected facilities in the confined space, it is possible to choose whether to arrange personnel for maintenance immediately, or continue to use the facilities until the defects in the confined space reach the level of maintenance. to a certain extent and then repaired.

The convolutional neural network can recognize images and optimize the accuracy of the recognition results through training. Through training, it can gradually get rid of human judgment and realize intelligence. Make the judgment more stable and fast.

In addition, the information collection device may also include a gas sensor for collecting gas information in areas that require safety monitoring and detection in the confined space, the safety monitoring information is gas information, and the safety monitoring information identification module is gas information. The identification module, the safety monitoring information identification method is a gas identification method, the gas information collected by the gas sensor is used on the gas identification module to identify and calculate the gas information by using the gas identification method, and then output the identification result. The gas sensor can be used to measure the type, concentration and composition of the gas, and can detect specific components in the gas, such as the detection of carbon monoxide gas, the detection of gas gas, the detection of gas, the detection of freon (R11, R12), etc., using The gas sensor installed on the drone collects gas information in the confined space. On the one hand, it detects safety issues such as gas pollution and gas leakage in the confined space. Safety during maintenance in confined spaces.

In Embodiment 2, this embodiment is basically the same as Embodiment 1, and for the sake of simplicity of expression, only the difference between its main structure and Embodiment 1 will be described. The unexplained parts in this embodiment are the same as in Embodiment 1, the difference is that: the first rod 113, the second rod 114, the third rod 115 and the fourth rod 116 are hinged end to end to form a rhombus The installation frame 112, and there is a limit protrusion 1131 provided on the base 111 between the inner corners of the first rod 113 and the second rod 114 and used to connect the first rod 113 and the second rod 114 The second rod 114 pulls in the opposite direction so that the first rod 113 and the second rod 114 are close to the first return spring 1132 of the limiting protrusion 1131, and the third rod 115 and the A second return spring 1133 is provided between the inner corners of the fourth rod 116 . Specifically, the first return spring and the second return spring are tension springs. When the restricted space safety detection drone is hit, especially when the corner connection part is subjected to an impact force, the two rhombus-shaped The installation frame is squeezed and deformed in the opposite direction, and the first return spring and the second return spring cooperate with the limit protrusion to pull the installation frame to a normal shape. Through the buffering effect of the spring, it can effectively reduce the Confined space safety detects the damage to the UAV body, especially the rotor, when the UAV is hit, and has a certain shock-absorbing effect.

In summary, the present application has the following beneficial effects through structural improvement:

1. UAVs are used instead of personnel to detect the situation in confined spaces, which greatly reduces the insecurity of personnel entering confined spaces, and the detection is also very convenient, and the timeliness and stability of the judgment results are enhanced. According to the degree of defects detected in the facilities in the confined space, it is possible to choose whether to immediately arrange personnel for maintenance or continue to use the facilities, because the degree of individual defects does not necessarily affect the work of the facilities, and it is better to wait until the defects reach a certain level before performing maintenance. It is conducive to improving the effective utilization of resources and making the safety monitoring work in confined spaces safer and smarter.

2. The lightweight anti-collision frame can rebound or rotate when it collides with an object, which can effectively prevent the UAV, especially the rotor, from being damaged. Flying in an unfamiliar and complex environment greatly reduces the incidence of "flying planes", ensures that drones can normally collect information in confined spaces and greatly reduces the difficulty of controlling drones.

3. The upper part of the lower-mounted rotor is connected to the output shaft of the coaxial brushless DC motor. Although the stability of the lower-mounted rotor has decreased, the maneuverability of the UAV is stronger, which is more conducive to its use in complex confined spaces. Avoid obstacles in time, improve the safety of the UAV, and make it easier for the camera to be installed above the UAV to capture a wider upper image.

4. A cylindrical frame is set between the first hemispherical frame and the second hemispherical frame, so that when the anti-collision frame falls to the ground, on the one hand, it can still offset part of the impact force by rolling, on the other hand, the round The cylindrical frame can roll in line contact with the ground, preventing the long-term unstable point contact rolling between the circular anti-collision frame and the ground, causing multi-directional and irregular flips on the drone body and increasing the chance of damage.

5. When the drone finds the faulty part, it can use a marking gun to mark the faulty part, such as marking an X-shaped or triangular pattern, or even a pattern with a fluorescent effect to make the mark more prominent. Of course, the mark can be used on the faulty part Or near the fault location, so that personnel can find the fault location more intuitively, accurately and quickly when performing maintenance on facilities in a confined space, and the positioning effect combined with the GPS module is better.

The foregoing is one or more implementation modes provided in conjunction with specific content, and it is not considered that the specific implementation of the application is limited to these descriptions. Anything that is similar or identical to the methods and structures of this application, or some technical deduction or replacement based on the concept of this application should be regarded as the scope of protection of this application.

Claims (10)

1. The restricted space safety detection method based on unmanned aerial vehicle is characterized in that, adopts the information acquisition device of restricted space safety detection UAV to carry out safety monitoring information collection to the area that needs safety monitoring detection in the restricted space, and all The collected safety monitoring information uses the safety monitoring information identification method to carry out the safety monitoring information identification operation on the safety monitoring information identification module, and then outputs the identification result. 2. the restricted space safety detection method based on unmanned aerial vehicle according to claim 1, is characterized in that, described information acquisition device is camera, and described safety monitoring information is image information, and described safety monitoring information identification module is Image recognition module, the safety monitoring information recognition method is an image recognition method, and the image information collected by the camera is used on the image recognition module to use the image recognition method to recognize and calculate the image information and then output the recognition result. The image recognition method includes the following step: S21, classify the detection scenes in the restricted space, and match the identification images with classified defects for each type of detection scene to make a training data set; S22, inputting the identification image that has classified the detection scene and defect degree in the training data set in step S21, and performing image sharpening and edge enhancement processing; S23, using the training data set formed by the identification image processed in step S22 to train the convolutional neural network; S24, input the image information collected in step S1 into the convolutional neural network trained in step S22, and output the recognition result. 3. the restricted space safety detection method based on unmanned aerial vehicle according to claim 2, is characterized in that, described in step S23, training convolutional neural network comprises the following steps: S231, performing the first convolution processing on the identification images in the training data set processed in step S22, forming the first layer of the convolutional neural network; S232: Perform the first pooling process on the first layer of the convolutional neural network, and perform local response normalization processing to form the second layer of the convolutional neural network; S233. Perform a second convolution process on the second layer of the convolutional neural network to form the third layer of the convolutional neural network; S234, performing a second pooling process on the third layer of the convolutional neural network to form the fourth layer of the convolutional neural network; S235. Obtain the fifth layer of the convolutional neural network in a fully connected manner for the fourth layer of the convolutional neural network; S236, the fifth layer of the convolutional neural network is fully connected to obtain an output layer: the output layer is divided into 6 independent parts, and each part is additionally connected to an independent loss function; S237. Minimize the value of the loss function by using the identified identification images that have been classified in S21, and complete the training of the convolutional neural network. 4. The method for safe detection of confined spaces based on unmanned aerial vehicles according to claim 2, wherein the detection scenarios are classified into gas pipeline burners, support storage tanks, inside industrial chimneys, inside large oil tankers, For a large pipeline or inside a coal-fired power plant boiler, the defect degree is classified into severe corrosion, moderate corrosion or mild corrosion, and the identification result is severe corrosion, moderate corrosion or mild corrosion. 5. The UAV-based restricted space safety detection method according to claim 3, characterized in that, the size of the identification image processed in step S22 is unified to 32 × 32 × 1, and the filter of the convolutional layer The size is 5×5, the depth is 6, all 0s are not used to supplement, and the step size is 1, the output size of the first layer of the convolutional neural network obtained is 32-5+1=28, and the depth is 6, that is, the The second layer node matrix of the convolutional neural network has 28×28×6=4704 nodes, and each node is connected to 5×5=25 current layer nodes, that is, the first layer of the convolutional neural network has a total of 4704×(25+ 1) = 122304 connections; The size of the filter used in the second layer of the convolutional neural network is 2×2, and the step size is 2, and the size of the output matrix of the second layer of the convolutional neural network is 14×14×6; The output matrix of the second layer of the convolutional neural network is the input of the third layer of the convolutional neural network. The size of the filter used is 5×5, the depth is 16, and all 0s are not used to supplement, the step size is 1, and the volume The output size of the third layer of the convolutional neural network is 14-5+1=10, and the depth is 16, that is, the size of the output matrix of the third layer of the convolutional neural network is 10×10×16, and the parameters of the third layer of the convolutional neural network are 5 ×5×6×16+16=2416, there are 10×10×16×(5×5+1)=41600 connections; The size of the input matrix of the fourth layer of the convolutional neural network is 10×10×16, the size of the filter used is 2×2, and the step size is 2, and the size of the output matrix of the fourth layer of the convolutional neural network is 5 ×5×16; The output matrix of the fourth layer of the convolutional neural network is the input of the fifth layer of the convolutional neural network, and it is straightened into a vector with a length of 5×5×16, that is, a three-dimensional matrix is pulled into a one-dimensional space Represented in the form of a vector, enter the fully connected layer for training, the number of output nodes of the fifth layer of the convolutional neural network is 120, that is, there are 5×5×16×120+120=48120 parameters in total. 6. The method for safe detection of confined spaces based on unmanned aerial vehicles according to claim 1, wherein said information collection device also includes a device for collecting the gas information of areas requiring safety monitoring and detection in confined spaces. For a gas sensor, the safety monitoring information is gas information, the safety monitoring information identification module is a gas identification module, the safety monitoring information identification method is a gas identification method, and the gas information collected by the gas sensor is used in the gas identification The module uses the gas identification method to identify and calculate the gas information and then outputs the identification results. 7. the restricted space safety detection method based on unmanned aerial vehicle according to claim 2, is characterized in that, described restricted space safety detection unmanned aerial vehicle comprises unmanned aerial vehicle body (1) and is located at described unmanned The anti-collision frame (2) outside the machine body (1); the drone body (1) includes a frame (11) and a cloud platform (12) installed above the frame, and the frame (11 ) is provided with a down-mounted rotor (110), the camera (121) is located on the platform (12), and one side of the camera (121) is provided with an illuminating light that moves with the camera (121) (122); The anti-collision frame (2) comprises a first hemispherical frame (21), a second hemispherical frame (22) and a connecting rod (23), the first hemispherical frame (21) and the second hemispherical frame Cutting edges of the frame (22) are connected to each other through the connecting rods (23) to form the anti-collision frame (2), a plurality of connecting rods (23) are arranged along the first hemispherical frame (21) and the second hemispherical frame (21) Cutting edges of the two hemispherical frames (22) are uniformly distributed to form a cylindrical frame connecting the first hemispherical frame (21) and the second hemispherical frame (22). 8. the restricted space safety detection method based on unmanned aerial vehicle according to claim 7, is characterized in that, described frame (11) comprises pedestal (111), and described pedestal (111) both sides symmetrical design There is an installation frame (112), and the installation frame (112) includes a first rod (113) and a second rod (113) that are connected to the base (111) and protrude from one side of the base (111) in a figure-eight shape. Two rods (114), a third rod (115) connected to the extended end of the first rod (113) and a fourth rod (116) connected to the extended end of the second rod (114), the The free ends of the third rod (115) and the fourth rod (116) intersect to form a corner connecting portion (117) connected with the anti-collision frame (2), and the lower-mounted rotor (110) is arranged on Below the connecting portion of the first rod (113) and the third rod (115) and below the connecting portion of the second rod (114) and the fourth rod (116); The first rod (113), the second rod (114), the third rod (115) and the fourth rod (116) are sequentially hinged end to end to form the diamond-shaped installation frame (112), and Between the inner corners of the first rod (113) and the second rod (114), there is a limit protrusion (1131) arranged on the base (111) and used to place the first rod (113) ) and the second rod (114) pull in the opposite direction to make the first rod (113) and the second rod (114) close to the first return spring ( 1132), a second return spring (1133) is provided between the inner corners of the third rod (115) and the fourth rod (116). 9. the restricted space safety detection method based on unmanned aerial vehicle according to claim 7, is characterized in that, described first hemispherical frame (21) comprises: Frame strips (211), which are provided with multiple pieces, one end of the many strips of frame strips (211) is gathered, and the other end is opened to form a hemispherical frame shape, and one end of the said frame strips (211) gathered is provided with the first hemispherical frame (21) a protrusion (2111) protruding from the top of the ball, and one end of the frame bar (211) opened is provided with a protrusion (2112) that is recessed toward the center of the cut surface of the first hemispherical frame (21); Gathering ring (212), which is arranged at one end of the frame bar (211) to gather, and said gathering ring (212) comprises annular hole (2121), annular gathering wall (2122) surrounding said annular hole (2121), Annular grooves (2123) evenly distributed along the annular retracting wall (2122) and outer ring threads (2124) provided on the outer wall of the annular wall (1142), the frame bar (211) is snapped into the annular groove (2123) and the said locking protrusion (2111) is folded in said annular hole (2121) so that one end of a plurality of said frame strips (211) forms a folded state; A screwing ring (213), which is provided with a fixing hole (2131) fixed to the frame (11), an annular screwing wall (2132) surrounding the fixing hole (2131), and an annular screwing wall provided on the ring (2132) the inner ring thread (2133) matched with the outer ring thread (2124) on the inner wall, and the gathering ring (212) is screwed into the screwing ring (213) after gathering the frame bar (211) Inside; Open the ring (214), which is evenly provided with clamping holes (2141) equal to the number of the frame bars (211) and lock holes (2142) equal to the number of the connecting rods (23) along the circumference, and the frame bars (211) The opened end is inserted into the clamping hole (2141); A connecting ring (215), which is uniformly provided with through holes equal to the number of connecting rods (23) along its circumference; A connecting screw, the connecting screw passes through the locking hole (2142) and the through hole to lock the expansion ring (214) and the connecting ring (215) on the connecting rod (23). 10. The UAV-based restricted space safety detection method according to any one of claims 7-9, characterized in that, the restricted space safety detection UAV is also provided with a real-time detection method for the restricted space. The position information of the space safety detection UAV is transmitted to the GPS module at the control end and the marking gun (3) used to mark the fault location in the confined space, and the marking gun (3) is an automatic spray gun capable of spraying colored solution.