CN109272497A

CN109272497A - Method for detecting surface defects of products, device and computer equipment

Info

Publication number: CN109272497A
Application number: CN201811037591.4A
Authority: CN
Inventors: 金玲玲; 饶东升; 何文玮
Original assignee: Shenzhen Lingtu Huishi Technology Co Ltd
Current assignee: Shenzhen Lingtu Huishi Technology Co Ltd
Priority date: 2018-09-05
Filing date: 2018-09-05
Publication date: 2019-01-25

Abstract

This application discloses method for detecting surface defects of products and device, computer equipment and computer-readable medium, this method includes obtaining image to be detected comprising product surface；Described image to be detected is detected using first nerves network model, the product surface for being included with the described image to be detected of determination is with the presence or absence of defect；Wherein, the first nerves network model carries out compression processing acquisition to trained nervus opticus network model using genetic algorithm, the nervus opticus network model is obtained using preset training sample training, and the first nerves network model is not less than default precision based on the precision of the training sample.This method and device, to product surface progress defects detection, have the advantages that calculation amount and memory space are lower, can be applied to the equipment stored and computing resource is all limited using by the neural network model after genetic algorithm compression processing.

Description

Product surface defect detection method and device and computer equipment

Technical Field

The present application relates to the field of computer application technologies, and in particular, to a method and an apparatus for detecting surface defects of a product, a computer device, and a computer readable medium.

Background

In recent years, with the development of artificial intelligence, Neural Network (NN) algorithms are widely applied to the field of product surface defect detection, such as fabric defect detection, electronic component surface defect detection, and the like. However, the deep neural network with a good detection effect often has a large number of nodes (neurons) and model parameters, which not only has a large calculation amount, but also occupies a large part of space in actual deployment, limiting the application of the deep neural network to devices with limited storage and calculation resources.

Disclosure of Invention

In view of the foregoing problems, embodiments of the present invention provide a method and an apparatus for detecting surface defects of a product, a computer device, and a computer readable medium, which are used for detecting surface defects of a product, and the neural network model has low computation amount and storage space, and can be applied to devices with limited storage and computation resources.

The method for detecting the surface defects of the product comprises the following steps: acquiring an image to be detected containing the surface of a product; detecting the image to be detected by utilizing a first neural network model so as to determine whether the surface of a product contained in the image to be detected has defects or not; the first neural network model is obtained by compressing a trained second neural network model by using a genetic algorithm, the second neural network model is obtained by training by using a preset training sample, and the precision of the first neural network model based on the training sample is not lower than the preset precision.

The product surface defect detection device according to the embodiment of the invention comprises: the acquisition module is used for acquiring an image to be detected containing the surface of a product; the detection module is used for detecting the image to be detected by utilizing a first neural network model so as to determine whether the surface of a product contained in the image to be detected has defects or not; the first neural network model is obtained by compressing a trained second neural network model by using a genetic algorithm, the second neural network model is obtained by training by using a preset training sample, and the precision of the first neural network model based on the training sample is not lower than the preset precision.

A computer apparatus according to an embodiment of the present invention includes: a processor; and a memory having executable instructions stored thereon, wherein the executable instructions, when executed, cause the processor to perform the foregoing method.

A computer-readable medium according to an embodiment of the present invention has executable instructions stored thereon, wherein the executable instructions, when executed, cause a computer to perform the aforementioned method.

From the above description, it can be seen that the scheme of the embodiment of the present invention performs defect detection on the product surface by using the first neural network model after the compression processing by the genetic algorithm, has the advantages of low calculation amount and storage space, and can be applied to devices with limited storage and calculation resources. Meanwhile, the scheme of the embodiment of the invention can simultaneously consider two aspects of detection accuracy and compression.

Drawings

FIG. 1 is a flow chart of a method of product surface defect detection according to one embodiment of the present invention;

FIG. 2 is a flow diagram of a method for compressing a trained second neural network model using a genetic algorithm, in accordance with one embodiment of the present invention;

FIG. 2a is an exemplary diagram of a neural network architecture;

FIG. 3 is a schematic view of a product surface defect detection apparatus according to one embodiment of the present invention;

FIG. 4 is a schematic diagram of a computer device in accordance with one embodiment of the present invention;

FIG. 5 is a block diagram of an exemplary computer device suitable for use to implement embodiments of the present invention, according to one embodiment of the invention.

Detailed Description

The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and thereby implement the subject matter described herein, and are not intended to limit the scope, applicability, or examples set forth in the claims. Changes may be made in the function and arrangement of elements discussed without departing from the scope of the disclosure. Various examples may omit, substitute, or add various procedures or components as needed. For example, the described methods may be performed in an order different from that described, and various steps may be added, omitted, or combined. In addition, features described with respect to some examples may also be combined in other examples.

As used herein, the term "include" and its variants mean open-ended terms in the sense of "including, but not limited to. The term "based on" means "based at least in part on". The terms "one embodiment" and "an embodiment" mean "at least one embodiment". The term "another embodiment" means "at least one other embodiment". The terms "first," "second," and the like may refer to different or the same object. Other definitions, whether explicit or implicit, may be included below. The definition of a term is consistent throughout the specification unless the context clearly dictates otherwise.

The embodiment of the invention adopts a first neural network model after the compression processing of the genetic algorithm to detect the surface of the product, and the genetic algorithm and the neural network are briefly introduced below.

Genetic Algorithm (GA) is a kind of randomized search method which is evolved by the evolution law of the biological world (survival of the fittest, and high-and-low-rejection Genetic mechanism). It was first proposed in 1975 by the teaching of j.holland in the united states, and its main feature is that the operation is directly performed on the structural object, and there is no derivation and function continuity limitation; the method has the advantages of inherent hidden parallelism and better global optimization capability; by adopting a probabilistic optimization method, the optimized search space can be automatically acquired and guided, the search direction can be adaptively adjusted, and a determined rule is not needed. These properties of genetic algorithms have been widely applied to fields such as combinatorial optimization, machine learning, signal processing, adaptive control, and artificial life. It is a key technology in modern related intelligent computing.

Neural Networks (NN) are a research hotspot in the field of artificial intelligence since the 80 th century of the 20 th century. The method abstracts the human brain neuron network from the information processing angle, establishes a certain simple model, and forms different networks according to different connection modes. A neural network is an operational model, which is formed by connecting a large number of nodes (or neurons). Each node represents a particular output function, called the excitation function. Each connection between two nodes represents a weighted value, called the connection weight, for the signal passing through the connection. The output of the network is different according to the connection mode, connection weight and excitation function of the network. The structural information of the neural network includes information such as nodes and connection rights.

FIG. 1 shows a flow diagram of a method for product surface defect detection according to one embodiment of the invention. The method 100 shown in fig. 1 may be implemented by a computer or other suitable electronic device having computing capabilities. Moreover, those skilled in the art will appreciate that any system that performs the method 100 is within the scope and spirit of embodiments of the present invention.

As shown in fig. 1, in step S102, an image to be detected including a surface of a product is acquired. In this embodiment, in specific implementation, the surface of the product to be detected may be photographed by a CCD industrial camera, and image data may be transmitted by a switch to obtain the image to be detected. The CCD (Charge Coupled Device) is a semiconductor Device for recording light changes in a digital camera.

In this embodiment, the size of the image to be detected may be 256 × 256. Of course, after the above-mentioned image to be inspected is obtained, the obtained image to be inspected may be subjected to image processing to obtain an image to be inspected having a size meeting the requirements.

In one embodiment, the product may be a fabric, a steel plate, glass, a magnetic tile, an electronic product, a workpiece, a plastic product, wood, or the like. Of course, it should be noted that the above-listed product types are only intended to better illustrate the embodiments of the present invention. In specific implementation, according to specific situations and detection requirements, the corresponding surface defects of other products besides the above-listed products can be detected selectively.

In step S104, detecting the image to be detected by using a first neural network model to determine whether the surface of a product contained in the image to be detected has defects; the first neural network model is obtained by compressing a trained second neural network model by using a genetic algorithm, the second neural network model is obtained by training by using a preset training sample, and the precision of the first neural network model based on the training sample is not lower than the preset precision.

In this embodiment, in a specific implementation, when an image to be detected is to be detected, preprocessing may be performed on the image to be detected, for example, but not limited to, converting the image to be detected into a grayscale image. Of course, the present invention is not limited to this, and in other embodiments of the present invention, in the case where the image to be detected is already suitable for detection using the model in the initial state, the image to be detected may not be preprocessed.

In this embodiment, in a specific implementation, when an image to be detected is to be detected, the image to be detected may be subjected to identification, positioning, and image segmentation processing, one or more candidate regions of the image to be detected are obtained through the identification, positioning, and image segmentation processing, and whether a product surface included in the candidate regions has a defect is determined by detecting the candidate regions. The identification localization and image segmentation process are known techniques, and a description thereof is omitted here.

The first neural network model for detecting the surface defects of the product is obtained by compressing the trained second neural network model by using a genetic algorithm. In the embodiment, the principle of performing compression processing on the second neural network model by using the genetic algorithm is based on the principle of "superior or inferior" of the genetic algorithm, and under the condition of considering the accuracy of the neural network model, the "compressed neural network model" is taken as a criterion to perform various genetic operations on the trained second neural network model, so as to finally obtain the first neural network model with a simplified structure. In specific implementation, a preset precision based on a preset training sample is set to constrain the compression process of the second neural network model, wherein the second neural network model is obtained by training with the preset training sample, and the preset precision can be the original precision of the second neural network model or a numerical value slightly lower than the original precision. The preset precision may be set manually or set by the electronic device based on a preset algorithm, and the preset precision may be adjusted according to actual needs, which is not limited in this respect. In specific implementation, the preset training sample can be used for training the compressed neural network model to obtain a first neural network model.

In some optional implementations of this embodiment, the compressing includes deleting at least one node of the neural network model to be compressed and its corresponding connection, and/or deleting at least one connection of the neural network model to be compressed, so as to reduce the network complexity of the neural network model to be compressed, that is, improve the network simplification of the neural network model to be compressed. Preferably, the compression process includes deleting at least one node and its corresponding connection of the hidden layer of the neural network model to be compressed, and/or deleting at least one connection of the hidden layer. Here, the hidden layer (also called hidden layer) refers to other layers except for the input layer (inputtlayer) and the output layer (output layer).

In some optional implementations of this embodiment, the training sample may include a plurality of normal images and a plurality of problem images of the product, where the normal images refer to images of the product surface without defects, and the problem images refer to images of the product surface with defects. Training a neural network model using training samples is a known technique, and a description thereof is omitted here.

In an embodiment of the method 100, the compressing the trained second neural network model by using a genetic algorithm includes: performing genetic operation on the chromosome individual corresponding to the second neural network model by taking the fitness value based on compression as a standard to generate a chromosome individual with the optimal fitness value; and training the neural network model corresponding to the chromosome individual with the optimal fitness value by using the training sample to obtain the first neural network model.

In this embodiment, the compression-based fitness value is a fitness value that can reflect a network simplification degree (or a network complexity), for example, the larger the fitness value is, the higher the network simplification degree is, that is, effective compression is realized; the smaller the fitness value, the lower the network simplification, i.e. no effective compression is achieved. When the neural network model is compressed by using a genetic algorithm, chromosome individuals with large fitness values can be selected to perform genetic operation, and finally, the chromosome individuals with the largest fitness value in the chromosome individuals generated in the Nth generation population are the optimal chromosome individuals. It should be noted that, in other embodiments of the present invention, the larger the fitness value is, the higher the network complexity is, that is, no effective compression is implemented; the smaller the fitness value is, the lower the network complexity is, that is, effective compression is realized, when a neural network model is compressed by using a genetic algorithm, chromosome individuals with small fitness values can be selected to execute genetic operation, and finally, the chromosome individuals with the minimum fitness values in the chromosome individuals generated in the Nth generation population are the optimal chromosome individuals.

In some optional implementations of this embodiment, the fitness value may be calculated using the following fitness function:

or

Wherein f (i, t) represents the fitness of the ith individual of the tth generation; e (i, t) represents the network error of the neural network model corresponding to the ith individual of the tth generation; h (i, t) represents the network simplification of the ith individual of the tth generation.

In some optional implementations of this embodiment, E (i, t) may be calculated using the following formula:

wherein,and the neural network models corresponding to the ith individuals of the tth generation respectively are based on the expected output value and the actual output value of the preset qth training sample.

H (i, t) can be calculated using the following formula:

wherein m (i, t) is the node number of the ith individual of the tth generation. The more simplified the network structure, the larger the network simplification value.

In the implementation mode, the network error E (i, t) is utilized to constrain the compression processing process of the neural network model to be compressed, and both precision and compression can be achieved. The smaller the network error E (i, t), the higher the accuracy of the neural network model after the compression process. The larger the network simplification value is, the more simplified the structure of the neural network model after the compression processing is. Therefore, in the present embodiment, the fitness value is larger for chromosome individuals with smaller network errors and larger network simplification degrees.

In this embodiment, the optimal network structure of the neural network model can be obtained by performing a decoding operation on the optimal chromosome individual. In some optional implementations of this embodiment, when training the first neural network model obtained after the compression process, fine-tuning (fine-tuning) may be performed on the first neural network model. Therefore, the neural network model slightly lower than the preset precision can be finely adjusted to meet the preset precision requirement.

From the above description, it can be seen that the scheme provided by the embodiment of the present invention performs defect detection on the product surface by using the first neural network model compressed by the genetic algorithm, has the advantages of low computation amount and storage space, and can be applied to devices with limited storage and computation resources. Meanwhile, the scheme of the embodiment of the invention can simultaneously consider two aspects of detection accuracy and compression.

Fig. 2 is a flow chart illustrating a method for performing genetic manipulation on a chromosomal individual corresponding to the second neural network model to generate a fitness-optimized chromosomal individual according to an embodiment of the present invention, where the method 200 shown in fig. 2 may be implemented by a computer or other suitable electronic device with computing capabilities. Moreover, those skilled in the art will appreciate that any system that performs the method 200 is within the scope and spirit of embodiments of the present invention.

As shown in fig. 2, in step S202, structure information of the trained second neural network model is obtained. And training the second neural network model on a training sample which is preset previously, wherein the training is carried out so that the precision of the second neural network model meets the preset precision. The second Neural Network model of the embodiment of the present invention may be a Convolutional Neural Network (CNN: Convolutional Neural Network) model, a Convolutional Neural Network (RCNN: regional based Convolutional Neural Network) model based on region information, a Recurrent Neural Network (RNN: Recurrent Neural Network) model, a Long Short-Term Memory model (LSTM: Long Short-Term Memory) or a Gated cyclic Unit (GRU: Gated Recurrent Unit), and may also be other types of Neural Network models or a cascaded Neural Network model in which a plurality of types of Neural networks are combined. The structural information of the neural network model comprises node information and connection weight information, and the network structure can be formed by a coupling momentExpressed by an array, e.g. an N × N matrix C ═ C_ij) N × N denotes a network structure with N nodes, where c_ijThe value of (d) represents the connection weight from node i to node j; c. C_ij0 means no connection from node i to node j; c. C_iiRepresenting the bias of node i.

In step S204, the second neural network model is encoded according to the structural information to obtain a chromosome. The structure of the neural network model needs to be expressed as individual chromosome codes of a genetic algorithm so as to be calculated by the genetic algorithm. In one embodiment, let the second neural network model have N neurons, and the input layer, hidden layer, and output layer nodes with serial numbers ranging from 1 to N can represent the neural network structure by an N × N matrix. Now, the neural network structure with 7 nodes shown in fig. 2a is taken as an example to illustrate the encoding method of the second neural network model according to the present embodiment. Table 1 shows the node connection relationship of the neural network structure, and in table 1, the element corresponding to (i, j) in the matrix represents the connection relationship from the ith node to the jth node. Because the embodiment of the invention does not involve the modification of the connection right of the second neural network model when the second neural network model is compressed, the embodiment represents the connection relation of the nodes in a form of 0, 1 and-1, wherein "0" represents no connection; "1" indicates that the connection weight is 1 and has an excitation (excitation) effect, and is shown by a solid line in FIG. 2 a; "-1" indicates that the connection weight is-1, with an inhibiting (inhibiting) effect, indicated by the dashed line in fig. 2 a. It can be seen that table 1 is equivalent to the structure shown in fig. 2 a.

Table 1, connection relationship of example neural network structure of the present embodiment

According to the node connection relationship shown in table 1, the codes of the neural network can be represented in the form of a numeric string consisting of 0, 1, -1, and the elements (3, 1) to (7, 6) are sequentially connected from left to right and from top to bottom to form the following chromosome codes:

in step S206, population initialization is performed to generate an initial population based on the chromosome. In specific implementation, the copy operation may be performed on chromosomes encoded by the neural network model to be compressed, a predetermined number of chromosome individuals may be randomly generated, and the set of chromosome individuals may be used as the initial population. The size of the initial population is determined by the population size M, which may be, for example, but not limited to, 10-100. After the copy operation is performed on the chromosome, a plurality of identical chromosomes are obtained, each chromosome is a chromosome individual, the set of the plurality of chromosome individuals is used as an initial population, and the initialization of the population is completed.

In step S208, fitness values for individual chromosomes in the population are calculated. In this embodiment, the fitness function may be calculated by using the following formula:

The calculation of E (i, t) and H (i, t) can be performed by the calculation formulas given in the embodiment shown in FIG. 1.

In this embodiment, the fitness function includes formula ① and formula ②, where formula ① is based on the fitness function of the network error, which reflects the accuracy of the neural network model, and formula ② is based on the fitness function of the network simplification, which reflects the compression of the neural network model.

In step S210, it is determined whether a termination condition is reached. The termination condition may include a preset threshold of the number of iterations or a set convergence condition. The number of iterations may be set to 500, for example, but not limited to, and it is determined that the termination condition is reached when the number of iterations reaches 500. The convergence condition may be set, for example, but not limited to, to determine that the termination condition is reached when the fitness value satisfies a certain condition. In one embodiment, the convergence condition may be set to the fitness value while satisfying the following condition:

wherein E is₀For presetting a network error value, H₀A network simplification value is preset.

In step S212, if the result of the determination in step S210 is that the termination condition is not met, then using fitness value as a standard, selecting chromosome individuals with part of fitness value meeting the requirement, performing genetic operations such as copying, crossing or mutation, and thus generating a new generation population, and then returning to step S208. the selection standard of this embodiment can use the following steps (1) to calculate the fitness value of each chromosome individual in the population based on precision by formula ①, then calculating a first selection probability of the individual being selected, and selecting a first chromosome individual according to the first selection probability, (2) to calculate the fitness value of each chromosome individual in the population based on compression by formula ②, then calculating a second selection probability of the individual being selected, and selecting a second chromosome individual from the first chromosome individuals selected in step (1) according to the second selection probability.

Wherein p (i, t) is the selection probability of the ith generation of the ith individual, f (i, t) is the fitness of the ith generation of the ith individual, and f (sum, t) is the total fitness of the population of the tth generation. The selection standard of the embodiment can be adopted to take account of the precision and the compression of the neural network model.

And performing a duplication, crossover or mutation operation on the selected individual chromosome, and if the individual chromosome is selected by adopting the selection standard, performing the duplication, crossover or mutation operation on the selected second individual chromosome. The copy operation refers to directly copying the selected parent chromosome individual from the current generation to the new generation individual without any change. The cross operation refers to randomly selecting two parent chromosome individuals from a population according to the selection method, and mutually replacing partial components of the two parent chromosome individuals to form a new offspring chromosome individual. The mutation operation is to randomly select a parent chromosome from the population according to the selection method, then randomly select a node on the expression of the parent chromosome as a mutation point, and change the value of the mutation point gene into another effective value to form a new child chromosome.

Whether or not a crossover operation occurs can be determined by the crossover probability P_cThe method is to randomly generate a random number P between 0 and 1, when P is less than or equal to P_cThe crossover operation occurs when P > P_cAnd no crossover occurs. Similarly, whether mutation occurs or not can be determined by the mutation probability P_mSince it is prior art, a description thereof will be omitted herein.

In this embodiment, when performing the crossover operation, a crossover point may be randomly selected in each parent chromosome, and the crossover point is hereinafter referred to as a crossover segment. After the first parent individual chromosome deletes the crossed section, the crossed section of the second parent individual chromosome is inserted into the crossed point, and the first child individual chromosome is generated. Similarly, after the second parent individual is deleted of its crossover, the crossover of the first parent individual is inserted at its crossover point to form a second child individual. In this case, if two selected parent chromosome individuals are the same, but the cross points are different, the generated child chromosome individuals are different, so that the inbreeding is effectively avoided, and the global search capability is improved.

In this embodiment, when performing the mutation operation, one of the following operations may be randomly adopted: (a) deleting at least one node and corresponding connection in the hidden layer of the neural network model; (b) deleting at least one connection in the neural network model hidden layer; (c) randomly repairing the deleted nodes or connections with a certain probability; (d) and adding hidden layer nodes and randomly generating corresponding connection weights. The nodes are always deleted before the nodes are added, the number of added nodes is not larger than the number of deleted nodes, meanwhile, the nodes are added only when the deleted nodes cannot generate a good filial generation, and the mutation operation can ensure that the method is always carried out in the direction of compressing the neural network model.

In step S214, if the determination result in step S210 is that the termination condition is reached, outputting the optimal chromosome individual to obtain the first neural network model after the compression processing. In this embodiment, the optimal chromosome individual may be set to satisfy the following conditions:

f₀＝max[H(i，t)]

alternatively, the optimum individual chromosome may be set to satisfy the following conditions at the same time:

wherein f is₀Fitness of the best individual chromosome, E₀For the predetermined net error value, H (i, t) is the net reduction degree of the ith generation of the ith individual.

FIG. 3 shows a schematic diagram of a product surface defect detection apparatus according to an embodiment of the invention. The apparatus 300 shown in fig. 3 corresponds to the method for detecting surface defects of products, and since the embodiment of the apparatus 300 is substantially similar to the embodiment of the method, the description is simple, and the relevant points can be referred to the partial description of the embodiment of the method. The apparatus 300 may be implemented by software, hardware or a combination of software and hardware, and may be installed in a computer or other suitable electronic devices with computing capabilities.

As shown in fig. 3, the apparatus 300 may include an acquisition module 302 and a detection module 304. The acquisition module 302 is used to acquire an image to be detected including a surface of a product. The detection module 304 is configured to detect the image to be detected by using a first neural network model to determine whether a surface included in the image to be detected has a defect; the first neural network model is obtained by compressing a trained second neural network model by using a genetic algorithm, the second neural network model is obtained by training by using a preset training sample, and the precision of the first neural network model based on the training sample is not lower than the preset precision.

In an embodiment of the apparatus 300, the apparatus 300 further includes a genetic operation module, configured to perform a genetic operation on the individual chromosome corresponding to the second neural network model using the fitness value based on the compression as a standard, so as to generate an individual chromosome with an optimal fitness value; and the training module is used for training the neural network model corresponding to the chromosome individual with the optimal fitness value by using the training sample so as to obtain the first neural network model.

In another embodiment of the apparatus 300, the genetic manipulation module comprises: an obtaining unit, configured to obtain structural information of the second neural network model; the coding unit is used for coding the second neural network model according to the structural information so as to obtain a chromosome; the initialization unit is used for carrying out population initialization according to the chromosome to generate an initial population; a calculating unit for calculating fitness values of individual chromosomes in the population; a judging unit for judging whether a termination condition is reached; a genetic operation unit, which is used for selecting chromosome individuals with partial fitness values meeting the requirements by taking the fitness values as a standard and executing copy, cross or mutation operations so as to generate a new generation of population if the termination condition is not met; and the output unit is used for outputting the chromosome individual with the optimal fitness value if the termination condition is reached.

In yet another embodiment of the apparatus 300, the computing unit is further configured to: respectively calculating the fitness values of the chromosome individuals based on precision and compression; accordingly, the genetic manipulation unit is further configured to: obtaining a first selection probability of chromosome individuals in the population according to the accuracy-based fitness value, selecting first chromosome individuals according to the first selection probability, obtaining a second selection probability of chromosome individuals in the population according to the compression-based fitness value, and selecting second chromosome individuals from the first chromosome individuals according to the second selection probability; performing a duplication, crossover, or mutation operation on the second individual chromosome, thereby generating a new generation population.

FIG. 4 shows a schematic diagram of a computer device according to an embodiment of the invention. As shown in fig. 4, the computer device 400 may include a processor 402 and a memory 404, wherein the memory 402 has stored thereon executable instructions that, when executed, cause the processor 402 to perform the method 100 shown in fig. 1 or the method 200 shown in fig. 2.

FIG. 5 illustrates a block diagram of an exemplary computer device suitable for use in implementing embodiments of the present invention. The computer device 500 shown in fig. 5 is only an example and should not bring any limitations to the functionality or scope of use of the embodiments of the present invention.

As shown in fig. 5, computer device 500 is implemented in the form of a general purpose computing device. The components of computer device 500 may include, but are not limited to: a processor 502, a system memory 504, and a bus 506 that couples various system components (including the processor 502 and the system memory 504).

Bus 506 represents one or more of any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, and a processor or local bus using any of a variety of bus architectures. By way of example, such architectures include, but are not limited to, Industry Standard Architecture (ISA) bus, micro-channel architecture (MAC) bus, enhanced ISA bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnect (PCI) bus.

Computer device 500 typically includes a variety of computer system readable media. Such media can be any available media that is accessible by computer device 500 and includes both volatile and nonvolatile media, removable and non-removable media.

The system memory 504 may include computer system readable media in the form of volatile memory, such as Random Access Memory (RAM)508 and/or cache memory 510. The computer device 500 may further include other removable/non-removable, volatile/nonvolatile computer system storage media. By way of example only, the storage system 512 may be used to read from and write to non-removable, nonvolatile magnetic media (not shown in FIG. 5, commonly referred to as a "hard drive"). Although not shown in FIG. 5, a magnetic disk drive for reading from and writing to a removable, nonvolatile magnetic disk (e.g., a "floppy disk") and an optical disk drive for reading from or writing to a removable, nonvolatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 506 by one or more data media interfaces. System memory 504 may include at least one program product having a set (e.g., at least one) of program modules that are configured to carry out the functions of the embodiments of the invention described above with respect to fig. 1 or 2.

A program/utility 514 having a set (at least one) of program modules 516 may be stored, for example, in system memory 504, such program modules 516 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which examples or some combination thereof may comprise an implementation of a network environment. Program modules 516 generally perform the functions and/or methods described above in connection with the embodiments of fig. 1 or 2 of the present invention.

The computer device 500 may also communicate with one or more external devices 600 (e.g., keyboard, pointing device, display 700, etc.), with one or more devices that enable a user to interact with the computer device 500, and/or with any devices (e.g., network card, modem, etc.) that enable the computer device 500 to communicate with one or more other computing devices. Such communication may occur through an input/output (I/0) interface 518. Moreover, computer device 500 may also communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN) and/or a public network, such as the Internet) via network adapter 520. As shown, the network adapter 520 communicates with the other modules of the computer device 500 via the bus 506. It should be appreciated that although not shown in the figures, other hardware and/or software modules may be used in conjunction with computer device 500, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, among others.

The processor 502 executes various functional applications and data processing by executing programs stored in the system memory 504, for example, to implement the product surface defect detection method shown in the above-described embodiment.

Embodiments of the present invention also provide a computer-readable medium having executable instructions stored thereon, wherein the executable instructions, when executed, cause a computer to perform the method 100 shown in fig. 1 or the method 200 shown in fig. 2.

The computer-readable media of this embodiment may include RAM508, and/or cache memory 510, and/or storage system 512 in system memory 504 in the embodiment described above in fig. 5.

With the development of technology, the propagation path of computer programs is no longer limited to tangible media, and the computer programs can be directly downloaded from a network or acquired by other methods. Accordingly, the computer-readable medium in the present embodiment may include not only tangible media but also intangible media.

The computer-readable medium of the present embodiments may take any combination of one or more computer-readable media. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

A computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Computer program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, apparatus, or computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

Embodiments of the present invention are described with reference to flowchart illustrations and/or block diagrams of methods, apparatus, and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

The detailed description set forth above in connection with the appended drawings describes exemplary embodiments but does not represent all embodiments that may be practiced or fall within the scope of the claims. The term "exemplary" used throughout this specification means "serving as an example, instance, or illustration," and does not mean "preferred" or "advantageous" over other embodiments. The detailed description includes specific details for the purpose of providing an understanding of the described technology. However, the techniques may be practiced without these specific details. In some instances, well-known structures and devices are shown in block diagram form in order to avoid obscuring the concepts of the described embodiments.

The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other variations without departing from the scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. The product surface defect detection method comprises the following steps:

acquiring an image to be detected containing the surface of a product;

detecting the image to be detected by utilizing a first neural network model so as to determine whether the surface of a product contained in the image to be detected has defects or not;

the first neural network model is obtained by compressing a trained second neural network model by using a genetic algorithm, the second neural network model is obtained by training by using a preset training sample, and the precision of the first neural network model based on the training sample is not lower than the preset precision.

2. The method of claim 1, wherein the compressing the trained second neural network model using the genetic algorithm comprises:

performing genetic operation on the chromosome individual corresponding to the second neural network model by taking the fitness value based on compression as a standard to generate a chromosome individual with the optimal fitness value;

and training the neural network model corresponding to the chromosome individual with the optimal fitness value by using the training sample to obtain the first neural network model.

3. The method of claim 2, wherein the performing genetic operations on the individual chromosomes corresponding to the second neural network model to generate the individual chromosomes with the optimal fitness value comprises:

obtaining structural information of the second neural network model;

encoding the second neural network model according to the structural information to obtain a chromosome;

performing population initialization according to the chromosome to generate an initial population;

calculating fitness values of individual chromosomes in the population;

judging whether a termination condition is reached;

if the end condition is not met, selecting chromosome individuals of which the fitness values meet the requirements by taking the fitness values as a standard, executing copying, crossing or mutation operations to generate a new generation of population, and then returning to the step of calculating the fitness values of the chromosome individuals in the population;

and if the termination condition is reached, outputting the chromosome individual with the optimal fitness value.

4. The method of claim 3, wherein said calculating fitness values for individual chromosomes in a population comprises:

respectively calculating the fitness values of the chromosome individuals based on precision and compression;

correspondingly, the method selects chromosome individuals with partial fitness values meeting the requirement by taking the fitness values as a standard, and performs replication, crossover or mutation operations, thereby generating a new generation population, and comprises the following steps:

obtaining a first selection probability of chromosome individuals in the population according to the accuracy-based fitness value, selecting first chromosome individuals according to the first selection probability, obtaining a second selection probability of chromosome individuals in the population according to the compression-based fitness value, and selecting second chromosome individuals from the first chromosome individuals according to the second selection probability; performing a duplication, crossover, or mutation operation on the second individual chromosome, thereby generating a new generation population.

5. Product surface defect detection device includes:

the acquisition module is used for acquiring an image to be detected containing the surface of a product;

the detection module is used for detecting the image to be detected by utilizing a first neural network model so as to determine whether the surface of a product contained in the image to be detected has defects or not;

6. The apparatus of claim 5, wherein the apparatus further comprises:

the genetic operation module is used for executing genetic operation on the chromosome individual corresponding to the neural network model to be compressed by taking the fitness value based on compression as a standard so as to generate the chromosome individual with the optimal fitness value;

and the training module is used for training the neural network model corresponding to the chromosome individual with the optimal fitness value by using the training sample so as to obtain the first neural network model.

7. The apparatus of claim 6, wherein the genetic manipulation module comprises:

an obtaining unit, configured to obtain structural information of the second neural network model;

the coding unit is used for coding the second neural network model according to the structural information so as to obtain a chromosome;

the initialization unit is used for carrying out population initialization according to the chromosome to generate an initial population;

a calculating unit for calculating fitness values of individual chromosomes in the population;

a judging unit for judging whether a termination condition is reached;

a genetic operation unit, which is used for selecting chromosome individuals with partial fitness values meeting the requirements by taking the fitness values as a standard and executing copy, cross or mutation operations so as to generate a new generation of population if the termination condition is not met;

and the output unit is used for outputting the chromosome individual with the optimal fitness value if the termination condition is reached.

8. The apparatus of claim 7, wherein the computing unit is further to:

accordingly, the genetic manipulation unit is further configured to:

9. A computer device, comprising:

a processor; and

a memory having executable instructions stored thereon, wherein the executable instructions, when executed, cause the processor to perform the method of any of claims 1-4.

10. A computer readable medium having executable instructions stored thereon, wherein the executable instructions, when executed, cause a computer to perform the method of any of claims 1-4.