CN112291547B - An impurity removal system with image recognition function - Google Patents

Abstract

The invention discloses an impurity removal system with image recognition function, which comprises a rejection device, an image sensor and a microcomputer controller; the rejection device comprises a vibrating groove arranged obliquely and capable of oscillating up and down, an intercepting net embedded in the bottom of the center of the vibrating groove, and the like. Control components, impurity removal pipeline embedded in the middle of the vibrating tank; the image sensor is arranged on the edge of the vibrating tank to collect images; the microcomputer controller sequentially performs vignetting correction processing, automatic gain processing and white balance processing on the acquired images. The large particles of impurities in the image will generate a control signal according to the recognition result and send it to the intercepting net control component; the vibration tank vibrates up and down to make the impurities in the vibration tank move to the middle, and the intercepting net control component controls the intercepting net to rise according to the control signal, and the vibration tank is moved up and down. The large particles of impurities in the middle of the tank are intercepted in the intercepting net and removed through the impurity removal pipeline, and the dust and impurities passing through the intercepting net are discharged through the conveying trough at the end of the vibrating tank. It can meet the needs of online impurity removal of single or multiple groups of impurity removal machines.

Description

An impurity removal system with image recognition function

technical field

The invention belongs to the technical field of dust removal, and in particular relates to an impurity removal system with an image recognition function.

Background technique

Air-delivered dust removal is currently the mainstream technology of tobacco dust removal in domestic tobacco companies. Air-delivered dust removal is to transport the soot and other wastes generated in the production process of cigarette machines through the impurities removal pipeline to the impurities removal machine through the negative pressure suction generated by the fan. One-step screening and cleaning work. During the high-speed operation of the cigarette machine, it is inevitable that a small amount of tobacco stems and soot lumps will enter the downstream cleaning machine along with the soot. These tobacco stems and soot lumps are not only easy to block the dust removal pipeline due to their large size, but also It will increase the workload of the cleaning machine and shorten the service life of the cleaning machine.

The traditional cigarette impurity removal system adopts a screening structure, that is, the soot and other wastes generated in the production process of the cigarette machine first enter the vibration tank of the terminal impurity remover through the negative pressure suction air through the dust removal pipe, and then pass through the repeated vibration tank of the impurity remover. Buffeting, drop the powdered soot, small particles of tobacco stems or small pieces of soot into the lower collection bag through the vibrating slot screen, and finally collect the powdered soot in the waste by the collection bag by negative pressure suction. All of them are sucked into the impurity removal machine, and the small particles of tobacco stems or small pieces of soot in the waste are sucked into the stem extraction machine, which is crushed by the stem extraction machine and then packed.

The impurity removal system using this structure has obvious drawbacks: First, the structure cannot guarantee the complete separation of soot, tobacco stems and other wastes. During simple separation, the small pieces of tobacco stems will be vibrated and dropped into the collection bag together, resulting in incomplete removal of impurities by the impurity remover, and even the hard tobacco stems will cause damage to the internal parts of the impurity remover. Second, this structure cannot ensure that the soot lumps can be effectively shaken off. Since most of the soot lumps are in the form of viscous lumps, the buffeting of the vibrating groove cannot effectively shake off the soot lumps. These soot lumps follow the tobacco stems. After entering the stemmer, it will stick to the inner wall of the stemmer, which will seriously affect the normal operation of the internal motor of the stemmer or block the stemmer channel. Third, this structure cannot ensure that the negative pressure suction in the cleaning machine remains smooth for a long time. The soot, tobacco stems and other wastes in the dust removal pipe are driven by the negative pressure suction and finally flow to the cleaning machine. Due to the vibration groove in the structure The screen cannot effectively separate the small pieces of tobacco stems and ash lumps. In the long run, the small pieces of tobacco stems and soot lumps will accumulate on the surface of the screen, hindering the flow of negative pressure suction near the screen, causing the negative pressure suction port to be blocked, resulting in The suction air is too small, and the negative pressure suction air there and the negative pressure suction air in the dust removal pipe network are one air source, which will jointly affect the smooth negative pressure suction air in the dust removal pipe, and it is easy to cause blockage in the dust removal pipe. Once blocked, it will consume a lot of human and material resources to break the pipe to remove the blockage.

Therefore, there is a need for an impurity removal system that can quickly detect and accurately remove large foreign objects. Referring to other similar materials in China, there are many methods proposed for this, such as: South China Normal University (Wei Yanda. Design and implementation of IC card surface printing defect detection system based on machine vision [D]. South China Normal University Master Thesis, 2015) A printing image defect detection scheme based on image difference discrimination method is designed. The scheme is divided into two steps: preprocessing and real-time detection. Preprocessing mainly corrects lens distortion, sets various image detection parameters and thresholds, and makes standard templates. The real-time detection comprehensively applies a variety of image processing algorithms such as median filtering, image grayscale, image registration, image difference, image segmentation, mathematical morphology processing, and Blob analysis. Yantai University (Zhao Junran. Research and Application of Glass Edge Grinding Defect Detection Based on Machine Vision [D]. Master Thesis of Shandong Yantai University, 2018) adopts machine vision technology to combine the imaging principle of industrial CCD camera, image processing technology (reduced noise, sharpening, segmentation, edge detection, mathematical morphology, and file storage format), and researched a feature extraction method for bright spots, burst edges and white lines in glass processing. Shaanxi University of Science and Technology (Wei Yiye. Research on Recognition of Paper Function Information Based on Image Processing Technology [D]. Master Thesis of Shaanxi University of Science and Technology, 2018) proposed an improved algorithm based on ORB, using BBF search algorithm to speed up the closest distance ratio matching criterion to After completing the matching and using the RANSAC algorithm to remove the mismatched points, the perspective transformation matrix is calculated to mark the identification. North China Electric Power University (Guo Tieqiao. Research on Barcode Recognition System Based on Image Processing Technology [D]. Master Thesis of North China Electric Power University, 2014) Using a digital camera with a CCD image sensor, the maximum inter-class method is used to identify the barcode image, and using The gamma correction algorithm corrects the image.

The above method is to improve the scanning accuracy by scanning the barcode and adopting a corresponding algorithm, and this method cannot be applied to cured products such as soot and tobacco stems. Or by adding light to the shooting scene, adding an image sensor, etc., to maximize the feature quantity of bright spots, burst edges and white lines in the image, so as to enhance the clarity and recognition of the acquired image. Or make differential adjustments by comparing standard settings or templates, which are fixed and less adaptable to the environment and scene. Therefore, these methods have certain limitations in identifying wastes such as flowing soot and tobacco stems.

SUMMARY OF THE INVENTION

In view of the above, the present invention provides an impurity removal system with an image recognition function, which scans and recognizes wastes such as soot and tobacco stems through an image sensor, and effectively removes small segments of tobacco stems and soot lumps through a rejecting device, which can meet the requirements of a single group Or multiple groups of impurity removal machines are required for online impurity removal.

The technical scheme of the present invention is:

An impurity removal system with image recognition function, comprising a removal device, an image sensor, and a signal processor;

The rejecting device includes a vibrating groove that is inclined and can oscillate up and down, an intercepting net and its control assembly embedded in the bottom of the center of the vibrating groove, and an impurity-removing pipeline embedded in the middle of the vibrating groove;

The image sensor is arranged on the edge of the vibrating groove and is used for collecting images;

The signal processor sequentially performs vignetting correction processing, automatic gain processing and white balance processing on the captured image, and then identifies large particles of impurities in the image, generates a control signal according to the identification result, and sends it to the interception network control component;

The vibrating tank vibrates up and down to move the impurities in the vibrating tank to the middle, and the intercepting net control component controls the lifting of the intercepting net according to the control signal, intercepts the large particles of impurities in the middle of the vibrating tank in the intercepting net, and removes impurities by removing impurities. The pipeline is removed, and the dust and impurities passing through the intercepting net are discharged through the conveying trough at the end of the vibrating tank.

Preferably, the intercepting net assembly includes a communication module, a jacking rod connected with the intercepting net, an air valve connected with the jacking rod, an air inlet pipe and an air outlet pipe connected with the air valve, and an air valve for controlling the expansion and compression of the air valve. The island, after receiving the control signal through the communication module, the air valve island controls the expansion of the air valve according to the control signal to drive the jacking rod to rise and then raise the intercepting net to realize the connection of large particles of impurities.

Preferably, the impurity removal pipeline includes a rejection port arranged at the bottom of the vibrating tank and a conveying line connected to the rejection port. The conveying line is provided with a valve, a filter screen and an air pressure regulating valve in sequence along the conveying direction, and the air pressure regulating valve is connected with a The negative pressure suction pipe is used to absorb the dust and impurities filtered by the filter screen. There is no dredging pipe on the conveying pipeline between the valve and the filter screen. The dredging pipe is equipped with negative pressure suction to absorb large particles of impurities.

Preferably, a valve motor is also provided on the conveying pipeline to control the opening or closing of the valve.

Preferably, the impurity removal system further includes a light source, and the light source may be a strobe light.

Preferably, at least one pair of vibrating groove feet is evenly distributed around the vibrating groove, and the vibrating groove base feet realize up and down oscillating motion through springs and then drive the vibrating groove to oscillate up and down. The image sensors are at least one pair, and are installed evenly on the edge of the vibrating groove.

Among them, the vignetting correction processing process for the acquired image is as follows:

First, establish a consistent white light source, and use the image sensor to collect the input intensity level of the white light source on the reference target with lower specular reflection. At this time, the image sensor points to the reference surface, records the response value of each pixel position, and then according to the following formula Calculate the correction factor for each pixel location:

Among them, I _REF (x, y) is the gray factor of the impurity at the pixel position (x, y), x and y tend to positive infinity integer values, J _LT (x, y) is the corresponding correction factor set, which stored in a lookup table;

Then, for the image captured by the image sensor at the same angle, the image pixel value is multiplied by the corresponding correction factor in the lookup table to achieve vignetting correction:

IMG(x,y)=H(x,y)*J _LT (x,y)

Among them, IMG(x, y) represents the image after vignetting correction, and H(x, y) represents the pixel value of the acquired image at the position (x, y).

Among them, the automatic gain processing process for the image after vignetting correction is as follows:

According to the irradiance and exposure system, the image after vignetting correction is processed by the following formula for automatic gain processing:

IMG'(x,y)=kL(x,y)+k(1-α)

Among them, IMG'(x,y) represents the image after automatic gain, L(x,y) is the irradiance value corresponding to the image IMG'(x,y), k represents the exposure coefficient when the image sensor collects the image (k is constant), α represents the resolution of the image sensor.

Among them, the white balance processing process of the image after automatic gain is as follows:

IMG' _BAL (x,y)=R' _BAL (x,y)*G' _BAL (x,y)*B' _BAL (x,y)

Among them, IMG' _BAL (x, y) represents the image after white balance, R' _BAL (x, y), G' _BAL (x, y), B' _BAL (x, y) respectively represent the RGB after white balance processing Three-channel image, the specific process is:

分别表示加权增益像素，其计算公式为：Among them, r represents the adjustment parameter, r∈[0,99], L(x _k ,y _k ) represents the irradiance value of the image IMG'(x,y) with the exposure coefficient k,

respectively represent the weighted gain pixels, and the calculation formula is:

Among them, k represents the exposure coefficient when the image sensor captures the image, R(x _k , y _k ) represents the R gain pixel with the exposure coefficient k, B(x _k , y _k ) represents the B gain pixel with the exposure coefficient k, G (x _k , y _k ) denotes the G-gain pixel with exposure coefficient k,

采用自适应加权增益像素

分别为：Preferably, during the white balance processing of the image after automatic gain, the weighted gain pixels are

adaptive weight gain pixel

They are:

Compared with the prior art, the beneficial effects of the present invention at least include:

The impurity removal system with the image recognition function provided by the embodiment of the present invention uses an image sensor to collect images, and sequentially performs vignetting correction processing, automatic gain processing, and white balance processing on the collected images through a processor. A control signal is generated, and the removing device realizes impurity removal according to the control signal, with high impurity removal efficiency and good effect.

Description of drawings

In order to illustrate the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative efforts.

1 is a schematic structural diagram of an impurity removal system with an image recognition function provided by an embodiment;

2 is a schematic structural diagram of an interception network and a control component thereof provided by an embodiment;

Fig. 3 is the closed state diagram of the top cover of the rejection opening provided by the embodiment;

Fig. 4 is the open state diagram of the top cover of the rejection opening provided by the embodiment;

5 is a schematic diagram of the screen structure provided by the embodiment;

6 is a schematic diagram of an image vignetting correction process provided by an embodiment;

7 is a schematic diagram of an automatic gain implementation process provided by an embodiment;

FIG. 8 is a schematic diagram of an image white balance process provided by an embodiment.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and do not limit the protection scope of the present invention.

FIG. 1 is a schematic structural diagram of an impurity removal system with an image recognition function provided by an embodiment. FIG. 2 is a schematic structural diagram of an interception network and a control component thereof provided by an embodiment. As shown in FIG. 1 and FIG. 2 , the impurity removal system provided by the embodiment includes: a strobe light 1 , a rejection device 2 , an image sensor 3 , a microcomputer controller 4 , a signal processor 5 , and a PLC controller 11 . Among them, the strobe light 1 uses an LED strobe light. When the image sensor 3 is filled with light, the flash duration of the LED strobe light is within 4ms. The flash duration of the fill light depends on the scene shooting pixel requirements, 1000 80ms for 10,000 pixels, 60ms for 8,000,000 pixels, and 28ms for 5,000,000 pixels, which can meet the requirements of various lighting conditions.

The rejecting device 2 includes a vibrating groove 6 that is inclined and presents an elliptical funnel shape, an intercepting net and its control assembly 7 embedded in the bottom of the center of the vibrating groove, an impurity-removing pipeline 8 embedded in the middle of the vibrating groove, a vibrating groove foot 9 and a vibrating groove. 6. The soot conveying trough 10 at the edge end, among which, there are 4 vibrating trough feet 9, which are evenly distributed around the vibrating trough 6, and the up and down oscillation movement is realized by the spring to drive the oscillating movement of the vibrating trough 6 up and down, and there are also 4 image sensors 3, It is evenly distributed and installed on the edge of the vibrating tank 6 for image acquisition.

As shown in FIG. 2 , the interception net and its control assembly 7 include an interception net 71 , a communication module 72 , a jacking rod 73 , an air valve 74 , an air intake pipe 75 , an air outlet pipe 76 , an air valve island 77 and a power switch 78 . The rod 73 is connected to the intercepting net, one end of the air valve 74 is connected to the jacking rod 73, the air inlet pipe 75 and the air outlet pipe 76 are connected to the other end of the air valve 74 at the same time, and the air valve island 76 is used to control the expansion and compression of the air valve, through the communication module 72 After receiving the control signal, the air valve island 77 controls the air valve 74 to expand according to the control signal to drive the jacking rod 73 to rise and then the intercepting net 71 to connect the large particles of impurities.

In the embodiment, the intercepting net 71 is an oval edged metal 200-mesh gauze, which is vibrated up and down by the positive pressure air below to block the lumps of soot and cigarette rods and gather them near the rejection port 81 .

The impurity removal pipeline 8 includes a rejection port 81 arranged at the bottom of the vibrating tank, and a conveying line 82 connected to the rejection port 81. The conveying line is provided with a valve 83, a filter screen 84, an air pressure regulating valve 85, and an air pressure regulating valve in sequence along the conveying direction. 85 is connected with a negative pressure suction pipe 86, which is used to absorb the dust and impurities filtered by the filter screen. There is no dredging pipe 87 on the conveying pipeline between the valve 83 and the filter screen 84. In order to absorb large particles of impurities, the conveying pipeline 82 is also provided with a valve motor 88 to control the opening or closing of the valve, and the opening or closing signal is uniformly controlled by the PLC controller 11 of the vibrating tank. The impurity removal pipeline 8 is mainly used to remove the large particles of impurities or dust such as block soot and tobacco rods through the negative pressure suction pipe 87 and send them into the collection bag. In addition, the loose soot is passed through another negative pressure suction pipe 86. Rejected and sent to other collection bags.

In this embodiment, the position of the front end of the removal port 81 can be adjusted and slightly protrudes, which can effectively absorb wastes such as soot and tobacco stems. And the rejection port 81 is supported by a top cover 89 through the top cover spring support, and the closed and open state diagrams of the top cover are shown in FIGS. 3 and 4 .

PLC controller 11 adopts Siemens S7-300 system controller, which has the characteristics of stable operation in high temperature and high humidity environment. Microcomputer controller 4 adopts Siemens 427D series microcomputer controller, which has the advantages of high operation reliability and rich interfaces. The captured image of the image sensor 3 is sent to the microcomputer controller 4 . The microcomputer controller 4 sequentially performs vignetting correction processing, automatic gain processing and white balance processing on the captured image to identify large particles in the image, and generates a control signal according to the identification result and sends it to the interception network control component 7 . The intercepting net control component 7 controls the lifting of the intercepting net 71 according to the control signal, intercepts the large particles of impurities in the middle of the vibration tank 6 in the intercepting net 71, and removes them through the impurity removal pipeline 8, and the dust and impurities passing through the intercepting net 71 pass through the vibration. The delivery chute at the end of the chute discharges.

In the impurity removal system, tobacco stems, soot and other wastes will be identified by the image sensor one by one before entering the impurity remover through the rejecting device, and the light source identified comes from the strobe light installed at the exit of the rejecting device, but the stroboscopic light The light emitted is diffusely emitted, and it cannot uniformly illuminate the tobacco stems and soot, which causes the attenuation of the light intensity when the light reaches the convex surface of the image sensor. This change with the position of tobacco stems and soot is called vignetting effect, which usually has the following influence factors: (1) Aperture effect: The aperture effect of the image sensor blocks part of the light, and the aperture effect is generated. Vignetting effect is more obvious. (2) Pupil aberration: The light diffusely emitted by the strobe light is partly refracted by surrounding objects (such as the wall of the dust removal tube), which will also result in a very uneven distribution of light passing through the aperture of the image sensor.

Vignetting correction should be performed due to the vignetting effect of the image sensor caused by the uneven distribution of the strobe light. Specifically, as shown in FIG. 6 , the vignetting correction processing process for the captured image is as follows:

First, establish a consistent white light source, and use the image sensor to collect the input intensity level of the white light source on the reference target with lower specular reflection. At this time, the image sensor points to the reference surface, records the response value of each pixel position, and then according to the following formula Calculate the correction factor for each pixel location:

Among them, I _REF (x, y) is the gray factor of the impurity at the pixel position (x, y), x and y tend to positive infinity integer values, J _LT (x, y) is the corresponding correction factor set, which stored in a lookup table;

Then, for the image captured by the image sensor at the same angle, the image pixel value is multiplied by the corresponding correction factor in the lookup table to achieve vignetting correction:

IMG(x,y)=H(x,y)*J _LT (x,y)

Among them, IMG(x, y) represents the image after vignetting correction, and H(x, y) represents the pixel value of the acquired image at the position (x, y).

By performing vignetting correction processing on the collected images, the vignetting effect generated during the image collection process can be weakened or eliminated, and the vignetting correction effect can be automatically realized, so that the images can achieve high-quality image collection in dark light scenes.

Wastes such as soot and tobacco stems flow all the time in the dust removal pipe, and the dynamic range of the scene captured by the image sensor is limited. Therefore, in the dynamic scene, compress the captured image to the maximum extent, and then expand the edge of the image. The automatic gain effect is realized, so that the collected images can achieve the largest range of collection under the high-speed flow state. As shown in Figure 7, when compressing the scene dynamic range of the image sensor, a nonlinear relationship is introduced between the pixel value H(x,y) at the image position (x,y) and the image irradiance L(x,y) , the nonlinear relationship can be described as:

H(x,y)=kL(x,y)

Among them, k represents the exposure coefficient when the image sensor acquires the image. Usually in the rejection device, when the waste such as soot and tobacco stems flow rapidly, the image sensor will easily lose frames after compressing the dynamic range of the acquired image, resulting in the deterioration of the image quality. , so it is necessary to perform automatic gain on the image.

Automatic gain is obtained by rapid gain changes in image sensor thermal imaging, assuming vignetting corrected images IMG ₁ (x,y) and IMG ₂ (x,y) at different exposure settings of the image sensor, by designing a The parameter model is used to characterize the consistency between the pixel values before and after the exposure change, so that a function related to the pixel values before and after the exposure change is generated, which is called the exposure response function. The exposure response function is constructed by the shot transformation, as follows:

C ₂ (x,y)=(choose(IMG ₁ (x,y),IMG ₂ (x,y)))

f(x,y)=compare(C ₁ (x,y),C ₂ (x,y))

where α represents the resolution of the image sensor, β represents the sensitivity of the image sensor, r represents the operating wavelength of the image sensor, C ₁ represents the pixel value of the image before exposure at position (x, y), and C ₂ represents the image at position (x, y) y) is the pixel value after exposure, f(x, y) represents the exposure response function, and outputs 0 or 1. 0 means that the image pixel does not change before and after exposure, and 1 means that the image pixel changes before and after exposure. C ₂ (x, y) means that the system will randomly select any one of the two images of IMG ₁ (x, y) or IMG ₂ (x, y) for the next comparison, and then compare the images after the comparison. Another image is compared, so use the choose command, so that only one image can be selected for each comparison, because the comparison between image pixels is very wasteful of computer memory and takes a lot of time, so only one image is used at a time. Comparing to save memory and time

For affine transformation, when the working wavelength of the image sensor is r=1, the exposure response function related to f(x,y) and f(kx,ky) is a linear function, as follows:

f(kx,ky)=kf(x,y)+α(1-k)

The working wavelength of the image sensor is a variable value, and the wavelength range is generally in the range of 0.2-1. Only when the wavelength of the image sensor reaches the maximum range, the exposure response function is linear, but the premise is that the image currently obtained by the image sensor must be gradual. The halo corrected image, that is, f(x,y)=chooseIMG(x,y), at this time, the exposure of the image sensor is the highest, that is, the automatic gain effect is the most obvious.

When the wavelength of the image sensor reaches the maximum range, the automatic gain effect of the image sensor is the most obvious, then the vignetting correction and automatic gain image of wastes such as soot and tobacco stems at the current position are:

IMG'(x,y)=kL(x,y)+k(1-α)

Among them, IMG'(x,y) represents the image after automatic gain, L(x,y) is the irradiance value corresponding to the image IMG'(x,y), k represents the exposure coefficient when the image sensor collects the image, and k is Constant, α represents the resolution of the image sensor.

If in the same scene, two images of soot and tobacco stems are continuously captured by the image sensor in the flowing state, and after vignetting correction and automatic gain, the gray levels of the two images have the following relationship:

IMG' ₂ (x,y)=T*IMG' ₁ (x,y)

In the formula, T is called the grayscale transfer function. Usually, T is a monotonically increasing function passing through the origin. The function value of T determines the smoothness and fineness of the image when it is converted, and the function value of T also determines the image. storage size. The automatic gain realization process is shown in Figure 7.

When the image sensor acquires images of wastes such as soot and tobacco stems, in addition to monochrome images, color images can also be used. Color images can more intuitively identify and scan the size, color and shape of waste, but the input color of color images is different, which is closely related to the strobe lighting conditions when the image is acquired. Different light sources have different spectral characteristics. Therefore, for the scene light source, it is necessary to adjust the acquired image to make the color of the image as true as possible. White balance technology is introduced for this purpose.

White balance is a method of automatically adjusting the color of the image. By looking for the off-white area in the image (the off-white area refers to the soot and tobacco stems that are acquired by the image sensor in a flowing state, and the acquired image has some areas of image color close to the real image color) to set the parameters, and then adjust the color of the rest of the image to approach or reach the color of the image in the near-white area.

Define IMG'(x,y) as an image that has undergone vignetting correction and automatic gain, and the size is M×N (M is the length, N is the width), then the white balance aims to adjust the color of IMG'(x,y) To generate a color-balanced RGB image, that is, a color image, generate an IMG' _BAL image, and its components are composed of R' _BAL (x, y), G' _BAL (x, y), B' _BAL (x, y), which is:

IMG' _BAL (x,y)=R' _BAL (x,y)*G' _BAL (x,y)*B' _BAL (x,y)

In the color image, the wavelength of the grayscale mode is the shortest, and the length of the wavelength directly determines the image storage size, so the grayscale mode of the RGB image (that is, the color image) is selected, so that the image quality and storage can be well balanced. size. After selecting the grayscale mode of the RGB image, when performing the image white balance, the white balance algorithm of the grayscale mode is used to generate R' _BAL (x,y), G' _BAL (x,y), B' by the following formulas _BAL (x,y), i.e.

分别表示加权增益像素，其计算公式为：Among them, r represents the adjustment parameter, r∈[0,99], L(x _k ,y _k ) represents the irradiance value of the image IMG'(x,y) with the exposure coefficient k,

respectively represent the weighted gain pixels, and the calculation formula is:

Among them, k represents the exposure coefficient when the image sensor captures the image, R(x _k , y _k ) represents the R gain pixel with the exposure coefficient k, B(x _k , y _k ) represents the B gain pixel with the exposure coefficient k, G (x _k , y _k ) denotes the G-gain pixel with exposure coefficient k,

均是彩色图像为灰度模式下进行运算求得，但是这种运算结果通常比较费时，搜索彩色图像中的类白区域需要花费较长时间。为此，需要将这种算法进行改进，使其运算结果可以全覆盖，从而提高搜索速度。这种改进算法称为自适应白平衡算法。Weighted Gain Pixels

All color images are obtained by operation in grayscale mode, but the result of this operation is usually time-consuming, and it takes a long time to search for white-like regions in color images. Therefore, this algorithm needs to be improved so that its operation results can be fully covered, thereby improving the search speed. This improved algorithm is called an adaptive white balance algorithm.

和

作任意变化，既可加权增益，也可加权渐晕，从而适应所有像素。彩色图像在灰度模式下的自适应白平衡算法为：The adaptive white balance algorithm is to replace the three pixels of R, G, and B with all pixels in the input image. In this algorithm, it is allowed to

and

Arbitrary changes can be made to weight both gain and vignetting to accommodate all pixels. The adaptive white balance algorithm for color images in grayscale mode is:

分别为RGB三通道自适应加权增益像素，相应的，在自适应白平衡算法中，通过下式生成R’_BAL(x,y)、G’_BAL(x,y)和B’_BAL(x,y)分别为：in,

They are RGB three-channel adaptive weighted gain pixels, correspondingly, in the adaptive white balance algorithm, R' _BAL (x, y), G' _BAL (x, y) and B' _BAL (x, y) are generated by the following formulas y) are:

In the formula, r is a constant value that can be adjusted to obtain the optimal result. For a 24-bit input image (color) in normal mode, r is selected as r∈[100,300], and for a 24-bit input image (color) in grayscale mode, The chosen value of r is r∈[0,99]. The image white balance process is shown in Figure 8. It can be seen that the r value selection is related to the image white balance effect. When selecting the r value, a corresponding value must be selected, rather than the larger the better, otherwise the image will be adaptively weighted. The gain will exceed the wavelength range, causing distortion.

The white balance processing of the image after vignetting correction and automatic gain processing can realize the acquisition of monochrome or color image of the sample, and automatically realize the white balance effect. Excellent and minimal image storage size.

The microcomputer controller sequentially performs vignetting correction processing, automatic gain processing and white balance processing on the captured image to identify large particles of impurities in the image, that is, tobacco stems, etc., and generates a control signal according to the recognition result and sends it to the interception network control component to control interception. The net 71 is raised to intercept large particles of impurities, and they are removed through the impurity removal pipeline.

The working process of the above-mentioned impurity removal system with image recognition function is as follows:

When impurities such as soot and tobacco stems enter the vibrating tank 6, the strobe light 1 starts to work, providing the lighting light source for shooting in the vibrating tank 6 and the need for supplementary light under various light conditions. A total of four image sensors 3 are installed around the vibrating slot 6 respectively, and are used to identify the lumpy soot and cigarette rods mixed in the soot. Since the vibrating slot 6 is inclined, a screen is arranged on the surface of the vibrating slot 6, which can be used for layering. Layer conveying, when the vibrating tank 6 moves up and down the block soot, tobacco rod and loose soot, and simply separate them. The surface of the mesh acts as a stepped structure (as shown in Figure 5), so it can be simply separated during the movement process. The precise separation has to be screened by subsequent image recognition, that is, when the vibration slot 6 passes through the four vibration slots at the bottom When the foot 9 moves up and down, the lumpy soot, tobacco rods and loose soot will gradually move from point B through the vibrating groove to point C and finally reach point A. During the process from point B to point C, the vibrating groove 6 is surrounded by The image sensor 3 will continuously collect images, and the image information is digitally processed by the signal processor 5 (vignetting correction processing, automatic gain processing and white balance processing) to generate a control signal and transmit it to the microcomputer controller 4. At this time, the microcomputer controller 4 executes The control strategy program that has been compiled by the control strategy method, after the program is executed, the screening information is output to the PLC controller 11, and the PLC controller 11 controls the intercepting net 71 at the lower end of the screen to make a jacking action to intercept the identified blocky soot and cigarette rods, at this time, these block soot and cigarette rods gather in the C area and can no longer move to point A, and the intercepted block soot and cigarette rods will gradually gather near the rejection port 81. When the negative pressure suction is turned on, Under the action of negative pressure suction, the top cover at the end of the rejection port 81 will be turned down, and these lumps of soot and cigarette rods will be sucked up by the rejection port. The negative pressure suction is closed, and the top cover of the rejection port is automatically turned upward by the spring support to ensure that the loose soot will not fall into the rejection port. And it is transported out through the soot conveying slot (the soot conveying slot also has negative pressure suction). The rejection port 81 adopts the shape of a three-tube round port, thereby increasing the suction range. When the lumps of soot and the cigarette rods go straight down through the valve 83 along the reject diameter, the valve 83 is controlled by the valve motor 88 to open and close. The lumps of soot and cigarette rods continue downward and pass through the filter screen 84 . At this time, the lumpy soot and the cigarette rods are intercepted by the filter screen 84 and sucked out of the pipe body by the dredging pipe 87 . The air pressure regulating valve 85 is used to adjust the negative pressure suction to an appropriate size, because too much suction will cause the filter screen 84 to wear too fast, and too small suction will cause blockage of soot and cigarette rods in the rejection port 81 . The suction pipe adopts a hose, and the end is connected to a vacuum pump.

In general, the vignetting correction steps provided by the embodiment are as follows: first, the strobe light will provide a white light source with a consistent and uniform light color, and the light source will ensure that wastes such as soot and cigarette rods can still be irradiated even in a dark environment. At this time, the image sensor installed around the vibrating tank will identify the block soot and cigarette rods mixed in these soot, and generate the pixel position, and then the system records the response value of each pixel position, and then converts to obtain each pixel position. A correction factor for each pixel position, and finally the subsequent images obtained by the image sensor at the same angle can be corrected by multiplying the image pixel value by the corresponding correction factor to achieve vignetting correction.

In general, the steps of automatic gain provided by the embodiment are: first, capture images of wastes such as soot, cigarette rods, etc. in a dynamic scene through an image sensor, and then perform vignetting correction by the system after the image is imaged, and then the system will automatically generate corresponding images. The exposure response function includes a wavelength range (change value) in the exposure response function. The value of the final wavelength range determines the linear relationship of the exposure response function, that is, determines whether the automatic gain effect of the system is obvious. In addition, the value of the wavelength range Size also determines the image storage size.

In general, the white balance steps provided by the embodiments are: first, capture images of wastes such as soot, cigarette rods, etc. in a dynamic scene through an image sensor, and after the image is imaged, the system performs vignetting correction and automatic gain successively, and then the system searches for A parameter is set for the off-white area in the image, and finally the system uses this parameter to adjust the color of the rest of the image to approach or reach the image color of the off-white area, even if the image color is close to the original color.

In this embodiment, the three processing procedures of vignetting correction, automatic gain and white balance are mutually dependent and inseparable, and the order of implementation of these three processing procedures cannot be reversed.

The above-mentioned impurity removal system with image recognition function uses an image sensor to collect images, and performs vignetting correction processing, automatic gain processing, and white balance processing on the collected images in sequence through a processor, after which medium and large particles of impurities are identified and a control signal is generated. , the removing device realizes the impurity removal according to the control signal, the impurity removal efficiency is high, and the effect is good.

The above-mentioned specific embodiments describe in detail the technical solutions and beneficial effects of the present invention. It should be understood that the above-mentioned embodiments are only the most preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, additions and equivalent substitutions made within the scope shall be included within the protection scope of the present invention.

Claims (9)

1. An impurity removal system with an image recognition function is characterized by comprising a rejection device, an image sensor and a signal processor;

the removing device comprises a vibration groove which is obliquely arranged and can vibrate up and down, an interception net and a control assembly thereof which are embedded in the bottom of the center of the vibration groove, and an impurity removing pipeline which is embedded in the middle of the vibration groove;

the image sensor is arranged at the edge of the vibration groove and used for collecting images;

the signal processor identifies large-particle impurities in the image after sequentially performing vignetting correction processing, automatic gain processing and white balance processing on the acquired image, generates a control signal according to an identification result and sends the control signal to the interception net control assembly;

the vibration tank vibrates up and down to enable impurities in the vibration tank to move towards the middle part, the interception net control assembly controls the interception net to rise according to a control signal, large-particle impurities in the middle part of the vibration tank are intercepted in the interception net and removed through an impurity removal pipeline, and dust impurities penetrating through the interception net are discharged through a conveying tank at the end part of the vibration tank;

the intercepting net assembly comprises a communication module, a lifting rod connected with the intercepting net, an air valve connected with the lifting rod, an air inlet pipe and an air outlet pipe connected with the air valve, and an air valve island for controlling expansion and compression of the air valve, wherein after the communication module receives a control signal, the air valve island controls expansion of the air valve according to the control signal to drive the lifting rod to rise and then rise the intercepting net so as to realize connection of large-particle impurities.

2. The system according to claim 1, wherein the impurity removing pipeline comprises a removing opening disposed at the bottom of the vibrating trough and a conveying pipeline connected to the removing opening, the conveying pipeline is sequentially provided with a valve, a filter screen and an air pressure regulating valve along a conveying direction, the air pressure regulating valve is connected to a negative pressure suction pipe for absorbing dust impurities filtered by the filter screen, the conveying pipeline between the valve and the filter screen is not provided with a dredging pipe, and the dredging pipe is provided with a negative pressure suction fan for absorbing large particle impurities.

3. The impurity removing system with the image recognition function as claimed in claim 2, wherein a valve motor is further arranged on the conveying pipeline and used for controlling the valve to be opened or closed.

4. The system according to claim 1, wherein at least 1 pair of feet of the vibration tank are uniformly distributed around the vibration tank, the feet of the vibration tank perform vertical oscillating motion through a spring to drive the vibration tank to perform vertical oscillating motion, and the image sensors are at least 1 pair and are uniformly distributed on the edge of the vibration tank.

5. The trash removal system with an image recognition function of claim 1 or 4, wherein the trash removal system further comprises a light source.

6. The impurity removing system with image recognition function according to claim 1, wherein the vignetting correction processing process for the collected image is as follows:

first, a uniform white light source is established, the input intensity level of the white light source on a specularly reflected reference target is captured with an image sensor, which is now directed at the reference surface, the response value for each pixel position is recorded, and then a correction factor for each pixel position is calculated according to the following formula:

wherein, I_REF(x, y) is the gray scale factor of the contaminant at the pixel location (x, y), x and y tend to be positive infinite integer values, J_LT(x, y) is a set of corresponding correction factors, which are stored in a look-up table;

then, aiming at the collected image of the image sensor at the same angle, the vignetting correction is realized by multiplying the pixel value of the image by the corresponding correction factor in the lookup table:

IMG(x,y)＝H(x,y)*J_LT(x,y)

where IMG (x, y) represents the vignetting corrected image, and H (x, y) represents the pixel value of the captured image at position (x, y).

7. The impurity removing system with image recognition function according to claim 1, wherein the automatic gain processing process of the image after vignetting correction is as follows:

and (3) performing automatic gain processing on the image after the vignetting correction according to the radiation illumination and the exposure system by adopting the following formula:

IMG'(x,y)＝kL(x,y)+k(1-α)

wherein, IMG '(x, y) represents the image after the automatic gain, L (x, y) is the radiance value corresponding to the image IMG' (x, y), k represents the exposure coefficient when the image sensor collects the image, k is a constant, and α represents the resolution of the image sensor.

8. The trash removal system with an image recognition function of claim 1, wherein the white balance processing of the image after the automatic gain is:

IMG'_BAL(x,y)＝R'_BAL(x,y)*G'_BAL(x,y)*B'_BAL(x,y)

wherein, IMG'_BAL(x, y) represents a white-balanced image, R'_BAL(x,y)、G'_BAL(x,y)、B'_BAL(x, y) are RGB three-channel images after white balance processing respectively, and the specific process is as follows:

wherein r represents an adjustment parameter, r is within 0,99]，L(x_k,y_k) Representing the radiance value of an image IMG' (x, y) containing an exposure coefficient k,

respectively representing weighted gain pixels, and the calculation formula is as follows:

where k denotes an exposure coefficient when the image sensor acquires an image, R (x)_k,y_k) Denotes an R gain pixel having an exposure coefficient k, B (x)_k,y_k) Representing a B gain pixel, G (x), having an exposure coefficient k_k,y_k) Representing the G gain pixel containing the exposure coefficient k,

9. the system of claim 8, wherein the weighted gain pixels are weighted during white balance processing of the automatically gained image

Using adaptively weighted gain pixels

Respectively as follows:

Images