Product mildewing grade evaluation method and system

Abstract

The invention discloses a product mildewing grade assessment method and system, and the method comprises the steps: obtaining a to-be-assessed mildewing image of a product, inputting the to-be-assessed mildewing image into a preset product region recognition model, and enabling the product region recognition model to predict the product region of the to-be-assessed mildewing image and the product pixel point information; inputting the to-be-evaluated mildewed image into a preset mildewed area identification model, so that the mildewed area identification model predicts a mildewed area and mildewed pixel point information of the to-be-evaluated mildewed image; determining product pixel points according to the product area and the product pixel point information of the to-be-evaluated mildewed image, and determining mildewed pixel points according to the mildewed area and the mildewed pixel point information of the to-be-evaluated mildewed image; and based on the product pixel points and the mildewed pixel points, determining the proportion of the mildewed area, and based on the proportion, determining a mildewed grade corresponding to the product. The mildewing grade of the product can be accurately and efficiently evaluated, and the applicability is wide.

Description

Technical Field

[0001] The present application relates to the technical field of three-proofing tests, and in particular to a method and system for evaluating the mold level of a product. Background Art

[0002] During the storage and transportation of products, mold is a frequent problem that may damage the appearance of the product, reduce its performance, or even cause the product to completely lose its function. By accurately assessing the mold level of the product, companies can quickly identify and resolve potential quality risks, ensuring that the product always meets the established quality standards and meets customer expectations. At the same time, this measure can also effectively prevent or reduce the occurrence of mold incidents, thereby reducing losses and improving overall operational efficiency.

[0003] The current mold level assessment method still relies heavily on the personal experience and subjective judgment of the testers, which may not only lead to inconsistency and accuracy of the assessment results, but also due to the lack of objective and quantitative assessment standards, the mold level determination results are difficult to be scientifically verified and repeated. Although some intelligent means have been used to determine the mold level, these methods are mainly limited to specific types of products. For products involving different material types and different process types, a unified, accurate and efficient mold level assessment scheme has not yet been formed.

[0004] Application Contents

[0005] The present application provides a method for evaluating the mold level of a product to accurately determine the proportion of a moldy area, thereby accurately and efficiently evaluating the mold level of a product, and has wide applicability.

[0006] In a first aspect, the present application provides a method for evaluating the mold grade of a product, comprising:

[0007] Acquire a moldy image of a product to be evaluated, and input the moldy image to be evaluated into a preset product area recognition model, so that the product area recognition model predicts the product area and product pixel information of the moldy image to be evaluated; wherein the product area recognition model is obtained by training based on a plurality of first product area images, and the plurality of first product area images are obtained by dividing moldy images of different products;

[0008] Inputting the moldy image to be evaluated into a preset moldy area recognition model, so that the moldy area recognition model predicts the moldy area and moldy pixel information of the moldy image to be evaluated; wherein the moldy area recognition model is obtained by training a plurality of first moldy area images, and the plurality of first moldy area images are obtained by dividing the second product area images determined by the product area recognition model;

[0009] Determine product pixels according to the product area and product pixel information of the moldy image to be evaluated, and determine moldy pixels according to the moldy area and moldy pixel information of the moldy image to be evaluated;

[0010] Based on the product pixels and the moldy pixels, the proportion of the moldy area is determined, and based on the proportion, the mold level corresponding to the product is determined.

[0011] The embodiment of the present application can quickly and accurately locate the product area and determine the product pixel information through the product area recognition model, and then accurately determine the product pixel in the moldy image, which can provide a direct basis for the subsequent evaluation of the mold level; by dividing the moldy images of different products, the first product area image can be accurately determined, which can avoid the interference of other area information on the model training, and facilitate the accuracy of the product area recognition model training; the moldy area recognition model can quickly and accurately locate the moldy area and determine the moldy pixel information, which can provide a direct basis for the subsequent evaluation of the mold level; using the trained product recognition model to accurately determine the second product area image, it can ensure that the subsequent moldy area recognition is only performed in the product area, avoiding the interference of background or other non-product areas, and improving the accuracy of the subsequent product mold level evaluation; based on the product pixel and the moldy pixel, the proportion of the moldy area can be accurately determined, and then based on the proportion, the mold level of the product can be quickly and accurately determined. Compared with the prior art, the present application can accurately determine the proportion of the moldy area, and then accurately and efficiently evaluate the mold level of the product, and has a wide range of applicability.

[0012] Furthermore, the method of enabling the moldy area recognition model to predict the moldy area and moldy pixel information of the moldy image to be evaluated is specifically as follows:

[0013] Determining first mask information corresponding to the moldy image to be evaluated based on the product area recognition model;

[0014] A first feature vector is extracted from the moldy image to be evaluated, a product area is determined based on the first mask information and the first feature vector, and product pixel information in the product area is extracted, wherein the first feature vector includes a first color feature vector, a first texture feature vector, and a first shape feature vector.

[0015] In this way, the product area recognition model can quickly and accurately locate the product area and determine the product pixel information, and then accurately determine the product pixels in the moldy image, which can provide a direct basis for the subsequent evaluation of the mold level.

[0016] Furthermore, the product area recognition model is obtained by training a plurality of first product area images, and the plurality of first product area images are obtained by dividing moldy images of different products, specifically:

[0017] Acquire moldy images of different products, divide the moldy images into a first training set and a second training set, and determine a first product area image in the first training set;

[0018] Training a first convolutional neural network based on the first product area image to obtain an initial product area recognition model;

[0019] Inputting the second training set into the initial product area recognition model to obtain a third product area image, and mixing the third product area image with the first product area image to obtain a fourth product area image;

[0020] The fourth product area image is interpolated to obtain a fifth product area image, product feature data of the fifth product area image is extracted, and the initial product area recognition model is trained based on the product feature data to obtain a product area recognition model.

[0021] In this way, by dividing the moldy images of different products obtained, the first product area image can be accurately determined, and the interference of other area information on model training can be avoided, which facilitates the subsequent efficient training of product area recognition models, and then facilitates the subsequent rapid and accurate positioning of product areas and determination of product pixel information.

[0022] Furthermore, the method of enabling the moldy area recognition model to predict the moldy area and moldy pixel information of the moldy image to be evaluated is specifically as follows:

[0023] Determining second mask information corresponding to the moldy image to be evaluated based on the moldy area recognition model;

[0024] A second feature vector is extracted from the moldy image to be evaluated, a moldy area is determined based on the second mask information and the second feature vector, and moldy pixel information in the moldy area is extracted, wherein the second feature vector includes a second color feature vector, a second texture feature vector, and a second shape feature vector.

[0025] In this way, the moldy area recognition model can quickly and accurately locate the moldy area and determine the moldy pixel information, which can provide a direct basis for the subsequent evaluation of the mold level.

[0026] Further, the moldy area recognition model is obtained by training a plurality of first moldy area images, and the plurality of first moldy area images are obtained by dividing the second product area images determined by the product area recognition model, specifically:

[0027] Determine a second product area image based on the trained product area recognition model, divide the second product area image into a third training set and a fourth training set, and determine a first moldy area image in the third training set;

[0028] Training a second convolutional neural network based on the first moldy area image to obtain an initial moldy area recognition model;

[0029] Inputting the fourth training set into the initial moldy area recognition model to obtain a second moldy area image, and mixing the second moldy area image with the first moldy area image to obtain a third moldy area image;

[0030] The third moldy area image is interpolated to obtain a fourth moldy area image, moldy feature data of the fourth moldy area image is extracted, and the initial moldy area recognition model is trained based on the moldy feature data to obtain a moldy area recognition model.

[0031] In this way, by dividing the second product area image, the first moldy area image corresponding to the product can be accurately determined, which facilitates the subsequent efficient training of the moldy area recognition model, and further facilitates the subsequent accurate positioning of the moldy area and determination of the moldy pixel point information.

[0032] In a second aspect, the present application provides a product mold grade assessment system, including: a first prediction module, a second prediction module, a determination module, and an assessment module;

[0033] The first prediction module is used to obtain a moldy image of a product to be evaluated, and input the moldy image to be evaluated into a preset product area recognition model, so that the product area recognition model predicts the product area and product pixel information of the moldy image to be evaluated; wherein the product area recognition model is obtained by training based on a plurality of first product area images, and the plurality of first product area images are obtained by dividing moldy images of different products;

[0034] The second prediction module is used to input the moldy image to be evaluated into a preset moldy area recognition model, so that the moldy area recognition model predicts the moldy area and moldy pixel information of the moldy image to be evaluated; wherein the moldy area recognition model is obtained by training based on a plurality of first moldy area images, and the plurality of first moldy area images are obtained by dividing the second product area images determined by the product area recognition model;

[0035] The determination module is used to determine product pixels according to the product area and product pixel information of the moldy image to be evaluated, and to determine moldy pixels according to the moldy area and moldy pixel information of the moldy image to be evaluated;

[0036] The evaluation module is used to determine the proportion of the moldy area based on the product pixels and the moldy pixels, and determine the mold level corresponding to the product based on the proportion.

[0037] The embodiment of the present application can quickly and accurately locate the product area and determine the product pixel information through the product area recognition model, and then accurately determine the product pixel in the moldy image, which can provide a direct basis for the subsequent evaluation of the mold level; by dividing the moldy images of different products, the first product area image can be accurately determined, which can avoid the interference of other area information on the model training, and facilitate the accuracy of the product area recognition model training; the moldy area recognition model can quickly and accurately locate the moldy area and determine the moldy pixel information, which can provide a direct basis for the subsequent evaluation of the mold level; using the trained product recognition model to accurately determine the second product area image, it can ensure that the subsequent moldy area recognition is only performed in the product area, avoiding the interference of background or other non-product areas, and improving the accuracy of the subsequent product mold level evaluation; based on the product pixel and the moldy pixel, the proportion of the moldy area can be accurately determined, and then based on the proportion, the mold level of the product can be quickly and accurately determined. Compared with the prior art, the present application can accurately determine the proportion of the moldy area, and then accurately and efficiently evaluate the mold level of the product, and has a wide range of applicability.

[0038] Further, the first prediction module includes: a first determination unit and a first extraction unit;

[0039] The first determining unit is used to determine first mask information corresponding to the moldy image to be evaluated based on the product area recognition model;

[0040] The first extraction unit is used to extract a first feature vector from the moldy image to be evaluated, determine a product area based on the first mask information and the first feature vector, and extract product pixel information in the product area, wherein the first feature vector includes a first color feature vector, a first texture feature vector, and a first shape feature vector.

[0041] Furthermore, the first prediction module further includes: a first division unit, a first training unit, a first mixing unit and a second training unit;

[0042] The first division unit is used to obtain moldy images of different products, divide the moldy images into a first training set and a second training set, and determine a first product area image in the first training set;

[0043] The first training unit is used to train a first convolutional neural network based on the first product area image to obtain an initial product area recognition model;

[0044] The first mixing unit is used to input the second training set into the initial product area recognition model to obtain a third product area image, and mix the third product area image with the first product area image to obtain a fourth product area image;

[0045] The second training unit is used to perform interpolation processing on the fourth product area image to obtain a fifth product area image, extract product feature data of the fifth product area image, and train the initial product area recognition model based on the product feature data to obtain a product area recognition model.

[0046] Furthermore, the second prediction module includes: a second determination unit and a second extraction unit:

[0047] The second determination unit is used to determine second mask information corresponding to the moldy image to be evaluated based on the moldy area recognition model;

[0048] The second extraction unit is used to extract a second feature vector from the moldy image to be evaluated, determine a moldy area based on the second mask information and the second feature vector, and extract moldy pixel information in the moldy area, wherein the second feature vector includes a second color feature vector, a second texture feature vector, and a second shape feature vector.

[0049] Furthermore, the second prediction module further includes: a second division unit, a third training unit, a second mixing unit and a fourth training unit;

[0050] The second division unit is used to determine a second product area image based on the trained product area recognition model, and divide the second product area image into a third training set and a fourth training set, and determine a first moldy area image in the third training set;

[0051] The third training unit is used to train the second convolutional neural network based on the first moldy area image to obtain an initial moldy area recognition model;

[0052] The second mixing unit is used to input the fourth training set into the initial moldy area recognition model to obtain a second moldy area image, and mix the second moldy area image with the first moldy area image to obtain a third moldy area image;

[0053] The fourth training unit is used to perform interpolation processing on the third moldy area image to obtain a fourth moldy area image, extract moldy feature data of the fourth moldy area image, and train the initial moldy area recognition model based on the moldy feature data to obtain a moldy area recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of an embodiment of a method for evaluating mold grade of a product provided by the present application;

[0055] Figure 2 It is a structural schematic diagram of an embodiment of a product mold grade evaluation system provided in the present application. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0057] It should be understood that the step numbers used in this article are only for the convenience of description and are not intended to limit the order in which the steps are executed.

[0058] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise.

[0059] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0060] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.

[0061] During product storage and transportation, mold is a frequent problem that may not only damage the appearance of the product and reduce its performance, but may even cause the product to completely lose its function. Accurate assessment of mold levels can help quickly identify and resolve quality risks, ensure product standards, and reduce losses. Currently, mold level assessments rely heavily on manual experience and lack objective quantitative standards, resulting in doubts about the accuracy of the assessment. Although intelligent means have been applied, they are limited to the mold level assessment of specific products. There is no unified and efficient mold assessment solution for products with different materials and processes.

[0062] Next, the nouns involved in this application are analyzed:

[0063] Color moment is an effective color feature proposed by Stricker and Orengo. This method uses the concept of moment in linear algebra to represent the color distribution in an image with its moment. The color distribution is described by the first-order color moment (average), the second-order color moment (variance), and the third-order color moment (skewness). Unlike the color histogram, the image description using color moment does not require quantification of image features. Since each pixel has three color channels in the color space, the color moment of the image has 9 components to describe it. Since the color moment has fewer dimensions, it is often used in combination with other image features.

[0064] The LBP method (Local binary patterns) is a method used for image feature classification in computer vision. The LBP method was first proposed by T.Ojala, M. inen, and D. Harwood proposed it for texture feature extraction. Later, the LBP method was used in conjunction with the HOG feature classifier to improve the detection effect on some data sets.

[0065] The Canny edge detection algorithm is based on a multi-order edge operator and was first proposed by John F. Canny in 1986. He not only gave the edge detection method, but also proposed the computational theory of edge detection. The Canny edge detector uses a detection model derived from the Gaussian model. Because the unprocessed image may contain noise, a Gaussian filter is applied to the original image. The result is a slightly smoothed image so that the global important parameters are not disturbed by a single noise pixel.

[0066] Based on this, the embodiments of the present application provide a method and system for evaluating the mold level of a product, which can accurately determine the proportion of moldy areas, and then accurately and efficiently evaluate the mold level of the product, and has wide applicability.

[0067] A method and system for evaluating the mold level of a product provided in an embodiment of the present application are specifically described through the following embodiments. First, a method for evaluating the mold level of a product in an embodiment of the present application is described.

[0068] A method for evaluating the mold level of a product provided in an embodiment of the present application relates to the field of three-proof test technology. A method for evaluating the mold level of a product provided in an embodiment of the present application can be applied to a terminal, can be applied to a server side, or can be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or can be configured as a server cluster or a distributed system composed of multiple physical servers, or can be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements a method for evaluating the mold level of a product, etc., but is not limited to the above forms.

[0069] The present application can also be used in numerous general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0070] Embodiment 1

[0071] Please refer to Figure 1 , Figure 1 It is a flow chart of an embodiment of a method for evaluating mold grade of a product provided by the present application, including steps S101 to S104;

[0072] Step S101, obtaining a moldy image of a product to be evaluated, and inputting the moldy image to be evaluated into a preset product area recognition model, so that the product area recognition model predicts the product area and product pixel information of the moldy image to be evaluated; wherein the product area recognition model is obtained by training based on a plurality of first product area images, and the plurality of first product area images are obtained by dividing moldy images of different products;

[0073] In some embodiments, the moldy image of the product to be evaluated is obtained, specifically, but not limited to, by a camera, a scanner or other image capturing device. For example, when a camera is used to obtain the moldy image to be evaluated, the industrial camera needs to be fixed above the test bench, and then the camera is adjusted to a suitable height, angle and brightness so that the moldy image to be evaluated can be clearly captured.

[0074] In some embodiments, before the moldy image to be evaluated is input into a preset product area recognition model, the moldy image to be evaluated is also preprocessed and normalized so that it can be input into the model for prediction.

[0075] In some embodiments, the product area recognition model is obtained by training based on several first product area images, and several of the first product area images are obtained by dividing moldy images of different products, including: obtaining moldy images of different products, and dividing the moldy images into a first training set and a second training set, and determining the first product area image in the first training set; training a first convolutional neural network based on the first product area image to obtain an initial product area recognition model; inputting the second training set into the initial product area recognition model to obtain a third product area image, and mixing the third product area image with the first product area image to obtain a fourth product area image; interpolating the fourth product area image to obtain a fifth product area image, extracting product feature data of the fifth product area image, and training the initial product area recognition model based on the product feature data to obtain a product area recognition model. Specifically, first, it is necessary to obtain moldy images of different products through a camera, scanner or other image capture equipment, and then the large model can be used to filter out overexposed or weak light images with poor training effects, and the different moldy images to be evaluated are classified according to product categories. When the classification is completed, the image needs to be normalized to unify the image size to obtain the preprocessed moldy image, and the preprocessed moldy image is divided into two parts, namely the first training set and the second training set. Then, the first training set is segmented at the pixel level to identify or mark the first product area image from the first training set; secondly, after the first product area image is obtained, the first product area image needs to be used to train a convolutional neural network (CNN) to learn how to identify the product area. After the training is completed, an initial product area recognition model is obtained; again, the initial product area recognition model is used to semi-automatically annotate the second training set to segment the third product area image and other area images in the second training set, and after checking and correcting the third product area image, it is mixed with the first product area image to obtain the fourth product area image; finally, bilinear interpolation processing is performed on several fourth product area images to obtain the fifth product area image, and then, the product feature data of the fifth product area image is extracted, and the initial product area recognition model is trained with the product feature data to obtain product area recognition models of different categories m = {m1, m2, m3, m4....mn}, where n is the number of categories, and each category corresponds to a product area recognition model.

[0076] It should be noted that the semi-automatic labeling of the second training set through the initial product area recognition model combines the advantages of manual and automatic labeling. The second training set detects the target through the previously preliminarily trained model, and finally uses manual labeling tools to check and improve it, which can improve the labeling efficiency and quality and reduce the burden of manual labeling.

[0077] It should be noted that bilinear interpolation processing includes upsampling and downsampling. Upsampling is used to enlarge the original image, and downsampling is used to reduce the original image, which facilitates higher accuracy in subsequent prediction of images of different sizes.

[0078] It should be noted that the product feature data includes product color features, product texture features and product shape features, among which the product color features are color moments, which can be extracted by but not limited to capturing the color distribution and statistical data of the fifth product area image; the product texture features are LBP features, which can be extracted by but not limited to grayscale histograms, edge direction histograms, spatial grayscale co-occurrence matrices, Tamura textures and other methods; the product shape can be extracted by but not limited to Canny operator edge detection.

[0079] In some embodiments, the steps of LBP feature extraction are as follows: first, the detection window is divided into 16×16 small areas (cells), and for a pixel in each cell, the 8 points in its annular neighborhood (or multiple points in the annular neighborhood) are compared clockwise or counterclockwise. If the central pixel value is larger than that of the neighboring point, the neighboring point is assigned a value of 1, otherwise it is assigned a value of 0, so that each point will obtain an 8-bit binary number (usually converted to a decimal number). Then calculate the histogram of each cell, that is, the frequency of occurrence of each number (assuming it is a decimal number) (that is, a binary sequence of statistics on whether each pixel point is larger than the point in the neighborhood), and then normalize the histogram. Finally, the statistical histograms of each cell are connected to obtain the LBP texture features of the entire image, and then SVM or other machine learning algorithms can be used for classification.

[0080] In this way, by dividing the moldy images of different products obtained, the first product area image can be accurately determined, and the interference of other area information on model training can be avoided, which facilitates the subsequent efficient training of product area recognition models, and then facilitates the subsequent rapid and accurate positioning of product areas and determination of product pixel information.

[0081] In some embodiments, before inputting the moldy image to be evaluated into a preset product area recognition model, the method further includes identifying the product category of the moldy image to be evaluated, and selecting a corresponding product area recognition model according to the product category.

[0082] In some embodiments, the product area recognition model is used to predict the product area and product pixel information of the moldy image to be evaluated. Specifically, first mask information corresponding to the moldy image to be evaluated is determined based on the product area recognition model; a first feature vector in the moldy image to be evaluated is extracted, the product area is determined based on the first mask information and the first feature vector, and the product pixel information in the product area is extracted, wherein the first feature vector includes a first color feature vector, a first texture feature vector and a first shape feature vector. Specifically, after the moldy image to be evaluated is input into the product area recognition model, the product area recognition model will divide the moldy image to be evaluated into SxS grids and output SxS first mask information, the first mask information identifies the position of the product area in the moldy image to be evaluated; then, the product area recognition model will extract the first color feature vector, the first texture feature vector and the first shape feature vector in the moldy image to be evaluated; finally, since the first mask information has identified the approximate position of the product area, it is necessary to combine the first feature vector to more accurately determine the boundary of the product area, and then output the product pixel point information corresponding to the product area through the model, wherein the product pixel point information includes pixel value and coordinate position, etc.

[0083] It should be noted that the first mask information is usually a binary image, in which the product area is white (or a specific color) and the non-product area is black (or another specific color).

[0084] It should be noted that the product area recognition model will output results in JSON format, which contains the pixel locations of the product areas of the corresponding level.

[0085] In this way, the product area recognition model can quickly and accurately locate the product area and determine the product pixel information, and then accurately determine the product pixels in the moldy image, which can provide a direct basis for the subsequent evaluation of the mold level.

[0086] Step S102, inputting the moldy image to be evaluated into a preset moldy area recognition model, so that the moldy area recognition model predicts the moldy area and moldy pixel information of the moldy image to be evaluated; wherein the moldy area recognition model is obtained by training a plurality of first moldy area images, and the plurality of first moldy area images are obtained by dividing the second product area images determined by the product area recognition model;

[0087] Furthermore, the moldy area recognition model is obtained by training based on a number of first moldy area images, and the several first moldy area images are obtained by dividing the second product area images determined by the product area recognition model, specifically: determining the second product area image based on the trained product area recognition model, and dividing the second product area image into a third training set and a fourth training set, and determining the first moldy area image in the third training set; training the second convolutional neural network based on the first moldy area image to obtain an initial moldy area recognition model; inputting the fourth training set into the initial moldy area recognition model to obtain a second moldy area image, and mixing the second moldy area image with the first moldy area image to obtain a third moldy area image; interpolating the third moldy area image to obtain a fourth moldy area image, extracting moldy feature data of the fourth moldy area image, and training the initial moldy area recognition model based on the moldy feature data to obtain a moldy area recognition model. Specifically, first, after the product area recognition model is trained, it is necessary to determine the second product area image based on the trained product area recognition model, and divide the second product area image into two parts, namely the third training set and the fourth training set, and then perform instance segmentation on the third training set to identify or mark the first moldy area image from the third training set; secondly, after the first moldy area image is obtained, it is necessary to use the first moldy area image to train a convolutional neural network (CNN) to learn how to identify the moldy area, and after the training is completed, an initial moldy area recognition model is obtained; thirdly, the initial moldy area recognition model is used to identify the moldy area. A second moldy area image in a fourth training set is identified, and after checking and correcting the second moldy area image, it is mixed with the first moldy area image to obtain a third moldy area image; finally, a plurality of third moldy area images are subjected to bilinear interpolation processing to obtain a fourth moldy area image, and then moldy feature data in the fourth moldy area image is extracted, and an initial moldy area recognition model is trained with the moldy feature data to obtain moldy area recognition models of different categories k={k1, k2, k3, k4....kn}, where n is the number of categories, and each category corresponds to a moldy area recognition model.

[0088] In this way, by dividing the second product area image, the first moldy area image corresponding to the product can be accurately determined, which facilitates the subsequent efficient training of the moldy area recognition model, and further facilitates the subsequent accurate positioning of the moldy area and determination of the moldy pixel point information.

[0089] Furthermore, the method of enabling the moldy area recognition model to predict the moldy area and moldy pixel information of the moldy image to be evaluated is specifically as follows: determining the second mask information corresponding to the moldy image to be evaluated based on the moldy area recognition model; extracting the second feature vector in the moldy image to be evaluated, determining the moldy area based on the second mask information and the second feature vector, and extracting the moldy pixel information in the moldy area, wherein the second feature vector includes a second color feature vector, a second texture feature vector and a second shape feature vector. Specifically, after the moldy image to be evaluated is input into the moldy area recognition model, the moldy area recognition model will output a number of second mask information, and the second mask information identifies the position of the moldy area in the moldy image to be evaluated. Subsequently, the model will extract the second color feature vector, the second texture feature vector and the second shape feature vector in the moldy image to be evaluated; finally, since the second mask information has identified the approximate position of the moldy area, it is necessary to combine the second feature vector to more accurately determine the boundary of the moldy area, and then output the moldy pixel point information corresponding to the moldy area through the model, wherein the moldy pixel point information includes pixel value and coordinate position, etc.

[0090] It should be noted that the moldy area recognition model will output results in JSON format, which contains the pixel locations of the moldy areas of the corresponding levels.

[0091] In this way, the moldy area recognition model can quickly and accurately locate the moldy area and determine the moldy pixel information, which can provide a direct basis for the subsequent evaluation of the mold level.

[0092] Step S103, determining product pixels according to the product area and product pixel information of the moldy image to be evaluated, and determining moldy pixels according to the moldy area and moldy pixel information of the moldy image to be evaluated;

[0093] In some embodiments, the array in the product JSON file is traversed, the number of elements with a value of 1 is counted, and the product pixel points obtained by counting are output; the array in the moldy JSON file is traversed, the number of elements with a value of 1 is counted, and the moldy pixel points obtained by counting are output.

[0094] Step S104, based on the product pixels and the moldy pixels, determining the proportion of the moldy area, and determining the mold level corresponding to the product based on the proportion.

[0095] In some embodiments, when the product pixel points and the moldy pixel points are obtained, the moldy pixel points are divided by the product pixel points to obtain the proportion of the moldy area. Then, the mold level is determined according to the proportion of the moldy area based on the national standard or other preset standards. For example, if the proportion is less than 25%, it is defined as level 1, and if the proportion is greater than 25 and less than 50%, it is defined as level 2.

[0096] The embodiment of the present application can quickly and accurately locate the product area and determine the product pixel information through the product area recognition model, and then accurately determine the product pixel in the moldy image, which can provide a direct basis for the subsequent evaluation of the mold level; by dividing the moldy images of different products, the first product area image can be accurately determined, which can avoid the interference of other area information on the model training, and facilitate the accuracy of the product area recognition model training; the moldy area recognition model can quickly and accurately locate the moldy area and determine the moldy pixel information, which can provide a direct basis for the subsequent evaluation of the mold level; using the trained product recognition model to accurately determine the second product area image, it can ensure that the subsequent moldy area recognition is only performed in the product area, avoiding the interference of background or other non-product areas, and improving the accuracy of the subsequent product mold level evaluation; based on the product pixel and the moldy pixel, the proportion of the moldy area can be accurately determined, and then based on the proportion, the mold level of the product can be quickly and accurately determined. Compared with the prior art, the present application can accurately determine the proportion of the moldy area, and then accurately and efficiently evaluate the mold level of the product, and has a wide range of applicability.

[0097] Embodiment 2

[0098] Please refer to Figure 2 , Figure 2 It is a structural schematic diagram of an embodiment of a product mold level evaluation system provided by the present application, comprising: a first prediction module 100, a second prediction module 200, a determination module 300 and an evaluation module 400;

[0099] The first prediction module 100 is used to obtain a moldy image of a product to be evaluated, and input the moldy image to be evaluated into a preset product area recognition model, so that the product area recognition model predicts the product area and product pixel information of the moldy image to be evaluated; wherein the product area recognition model is obtained by training based on a plurality of first product area images, and the plurality of first product area images are obtained by dividing moldy images of different products;

[0100] In some embodiments, the moldy image of the product to be evaluated is obtained, specifically, but not limited to, by a camera, a scanner or other image capturing device. For example, when a camera is used to obtain the moldy image to be evaluated, the industrial camera needs to be fixed above the test bench, and then the camera is adjusted to a suitable height, angle and brightness so that the moldy image to be evaluated can be clearly captured.

[0101] In some embodiments, before the moldy image to be evaluated is input into a preset product area recognition model, the moldy image to be evaluated is also preprocessed and normalized so that it can be input into the model for prediction.

[0102] In some embodiments, the first prediction module 100 includes: a first division unit, a first training unit, a first mixing unit and a second training unit; the first division unit is used to obtain moldy images of different products, and divide the moldy images into a first training set and a second training set, and determine a first product area image in the first training set; the first training unit is used to train a first convolutional neural network based on the first product area image to obtain an initial product area recognition model; the first mixing unit is used to input the second training set into the initial product area recognition model to obtain a third product area image, and mix the third product area image with the first product area image to obtain a fourth product area image; the second training unit is used to interpolate the fourth product area image to obtain a fifth product area image, extract product feature data of the fifth product area image, and train the initial product area recognition model based on the product feature data to obtain a product area recognition model. Specifically, first, it is necessary to obtain moldy images of different products through a camera, scanner or other image capture equipment, and then the large model can be used to filter out overexposed or weak light images with poor training effects, and the different moldy images to be evaluated are classified according to product categories. When the classification is completed, the image needs to be normalized to unify the image size to obtain the preprocessed moldy image, and the preprocessed moldy image is divided into two parts, namely the first training set and the second training set. Then, the first training set is segmented at the pixel level to identify or mark the first product area image from the first training set; secondly, after the first product area image is obtained, the first product area image needs to be used to train a convolutional neural network (CNN) to learn how to identify the product area. After the training is completed, an initial product area recognition model is obtained; again, the initial product area recognition model is used to semi-automatically annotate the second training set to segment the third product area image and other area images in the second training set, and after checking and correcting the third product area image, it is mixed with the first product area image to obtain the fourth product area image; finally, bilinear interpolation processing is performed on several fourth product area images to obtain the fifth product area image, and then, the product feature data of the fifth product area image is extracted, and the initial product area recognition model is trained with the product feature data to obtain product area recognition models of different categories m = {m1, m2, m3, m4....mn}, where n is the number of categories, and each category corresponds to a product area recognition model.

[0103] It should be noted that the semi-automatic labeling of the second training set through the initial product area recognition model combines the advantages of manual and automatic labeling. The second training set detects the target through the previously preliminarily trained model, and finally uses manual labeling tools to check and improve it, which can improve the labeling efficiency and quality and reduce the burden of manual labeling.

[0104] It should be noted that bilinear interpolation processing includes upsampling and downsampling. Upsampling is used to enlarge the original image, and downsampling is used to reduce the original image, which facilitates higher accuracy in subsequent prediction of images of different sizes.

[0105] It should be noted that the product feature data includes product color features, product texture features and product shape features, among which the product color features are color moments, which can be extracted by but not limited to capturing the color distribution and statistical data of the fifth product area image; the product texture features are LBP features, which can be extracted by but not limited to grayscale histograms, edge direction histograms, spatial grayscale co-occurrence matrices, Tamura textures and other methods; the product shape can be extracted by but not limited to Canny operator edge detection.

[0106] In some embodiments, the steps of LBP feature extraction are as follows: first, the detection window is divided into 16×16 small areas (cells), and for a pixel in each cell, the 8 points in its annular neighborhood (or multiple points in the annular neighborhood) are compared clockwise or counterclockwise. If the central pixel value is larger than that of the neighboring point, the neighboring point is assigned a value of 1, otherwise it is assigned a value of 0, so that each point will obtain an 8-bit binary number (usually converted to a decimal number). Then calculate the histogram of each cell, that is, the frequency of occurrence of each number (assuming it is a decimal number) (that is, a binary sequence of statistics on whether each pixel point is larger than the point in the neighborhood), and then normalize the histogram. Finally, the statistical histograms of each cell are connected to obtain the LBP texture features of the entire image, and then SVM or other machine learning algorithms can be used for classification.

[0107] In this way, by dividing the moldy images of different products obtained, the first product area image can be accurately determined, and the interference of other area information on model training can be avoided, which facilitates the subsequent efficient training of product area recognition models, and then facilitates the subsequent rapid and accurate positioning of product areas and determination of product pixel information.

[0108] In some embodiments, before inputting the moldy image to be evaluated into a preset product area recognition model, the method further includes identifying the product category of the moldy image to be evaluated, and selecting a corresponding product area recognition model according to the product category.

[0109] In some embodiments, the first prediction module 100 includes: a first determination unit and a first extraction unit; the first determination unit is used to determine the first mask information corresponding to the moldy image to be evaluated based on the product area recognition model; the first extraction unit is used to extract the first feature vector in the moldy image to be evaluated, determine the product area based on the first mask information and the first feature vector, and extract the product pixel point information in the product area, wherein the first feature vector includes a first color feature vector, a first texture feature vector and a first shape feature vector. After the moldy image to be evaluated is input into the product area recognition model, the product area recognition model will divide the moldy image to be evaluated into SxS grids and output SxS first mask information, the first mask information identifies the position of the product area in the moldy image to be evaluated; then, the product area recognition model will extract the first color feature vector, the first texture feature vector and the first shape feature vector in the moldy image to be evaluated; finally, since the first mask information has identified the approximate position of the product area, it is necessary to combine the first feature vector to more accurately determine the boundary of the product area, and then output the product pixel point information corresponding to the product area through the model, wherein the product pixel point information includes pixel value and coordinate position, etc.

[0110] It should be noted that the first mask information is usually a binary image, in which the product area is white (or a specific color) and the non-product area is black (or another specific color).

[0111] It should be noted that the product area recognition model will output results in JSON format, which contains the pixel locations of the product areas of the corresponding level.

[0112] In this way, the product area recognition model can quickly and accurately locate the product area and determine the product pixel information, and then accurately determine the product pixels in the moldy image, which can provide a direct basis for the subsequent evaluation of the mold level.

[0113] The second prediction module 200 is used to input the moldy image to be evaluated into a preset moldy area recognition model, so that the moldy area recognition model predicts the moldy area and moldy pixel information of the moldy image to be evaluated; wherein the moldy area recognition model is obtained by training a plurality of first moldy area images, and the plurality of first moldy area images are obtained by dividing the second product area images determined by the product area recognition model;

[0114] In some embodiments, the second prediction module 200 includes: a second division unit, a third training unit, a second mixing unit and a fourth training unit; the second division unit is used to determine the second product area image based on the trained product area recognition model, and divide the second product area image into a third training set and a fourth training set to determine the first moldy area image in the third training set; the third training unit is used to train the second convolutional neural network based on the first moldy area image to obtain an initial moldy area recognition model; the second mixing unit is used to input the fourth training set into the initial moldy area recognition model to obtain a second moldy area image, and mix the second moldy area image with the first moldy area image to obtain a third moldy area image; the fourth training unit is used to interpolate the third moldy area image to obtain a fourth moldy area image, extract moldy feature data of the fourth moldy area image, and train the initial moldy area recognition model based on the moldy feature data to obtain a moldy area recognition model. Specifically, first, after the product area recognition model is trained, it is necessary to determine the second product area image based on the trained product area recognition model, and divide the second product area image into two parts, namely the third training set and the fourth training set, and then perform instance segmentation on the third training set to identify or mark the first moldy area image from the third training set; secondly, after the first moldy area image is obtained, it is necessary to use the first moldy area image to train a convolutional neural network (CNN) to learn how to identify the moldy area, and after the training is completed, an initial moldy area recognition model is obtained; thirdly, the initial moldy area recognition model is used to identify the moldy area. A second moldy area image in a fourth training set is identified, and after checking and correcting the second moldy area image, it is mixed with the first moldy area image to obtain a third moldy area image; finally, a plurality of third moldy area images are subjected to bilinear interpolation processing to obtain a fourth moldy area image, and then moldy feature data in the fourth moldy area image is extracted, and an initial moldy area recognition model is trained with the moldy feature data to obtain moldy area recognition models of different categories k={k1, k2, k3, k4....kn}, where n is the number of categories, and each category corresponds to a moldy area recognition model.

[0115] In this way, by dividing the second product area image, the first moldy area image corresponding to the product can be accurately determined, which facilitates the subsequent efficient training of the moldy area recognition model, and further facilitates the subsequent accurate positioning of the moldy area and determination of the moldy pixel point information.

[0116] Furthermore, the second prediction module 200 includes: a second determination unit and a second extraction unit: the second determination unit is used to determine the second mask information corresponding to the moldy image to be evaluated based on the moldy area recognition model; the second extraction unit is used to extract the second feature vector in the moldy image to be evaluated, determine the moldy area based on the second mask information and the second feature vector, and extract the moldy pixel information in the moldy area, wherein the second feature vector includes a second color feature vector, a second texture feature vector and a second shape feature vector. Specifically, after the moldy image to be evaluated is input into the moldy area recognition model, the moldy area recognition model will output a number of second mask information, and the second mask information identifies the position of the moldy area in the moldy image to be evaluated. Subsequently, the model will extract the second color feature vector, the second texture feature vector and the second shape feature vector in the moldy image to be evaluated; finally, since the second mask information has identified the approximate position of the moldy area, it is necessary to combine the second feature vector to more accurately determine the boundary of the moldy area, and then output the moldy pixel point information corresponding to the moldy area through the model, wherein the moldy pixel point information includes pixel value and coordinate position, etc.

[0117] It should be noted that the moldy area recognition model will output results in JSON format, which contains the pixel locations of the moldy areas of the corresponding levels.

[0118] In this way, the moldy area recognition model can quickly and accurately locate the moldy area and determine the moldy pixel information, which can provide a direct basis for the subsequent evaluation of the mold level.

[0119] The determination module 300 is used to determine product pixels according to the product area and product pixel information of the moldy image to be evaluated, and to determine moldy pixels according to the moldy area and moldy pixel information of the moldy image to be evaluated;

[0120] In some embodiments, the array in the product JSON file is traversed, the number of elements with a value of 1 is counted, and the product pixel points obtained by counting are output; the array in the moldy JSON file is traversed, the number of elements with a value of 1 is counted, and the moldy pixel points obtained by counting are output.

[0121] The evaluation module 400 is used to determine the proportion of the moldy area based on the product pixels and the moldy pixels, and determine the mold level corresponding to the product based on the proportion.

[0122] In some embodiments, when the product pixel points and the moldy pixel points are obtained, the moldy pixel points are divided by the product pixel points to obtain the proportion of the moldy area. Then, the mold level is determined according to the proportion of the moldy area based on the national standard or other preset standards. For example, if the proportion is less than 25%, it is defined as level 1, and if the proportion is greater than 25 and less than 50%, it is defined as level 2.

[0123] The embodiment of the present application can quickly and accurately locate the product area and determine the product pixel information through the product area recognition model, and then accurately determine the product pixel in the moldy image, which can provide a direct basis for the subsequent evaluation of the mold level; by dividing the moldy images of different products, the first product area image can be accurately determined, which can avoid the interference of other area information on the model training, and facilitate the accuracy of the product area recognition model training; the moldy area recognition model can quickly and accurately locate the moldy area and determine the moldy pixel information, which can provide a direct basis for the subsequent evaluation of the mold level; using the trained product recognition model to accurately determine the second product area image, it can ensure that the subsequent moldy area recognition is only performed in the product area, avoiding the interference of background or other non-product areas, and improving the accuracy of the subsequent product mold level evaluation; based on the product pixel and the moldy pixel, the proportion of the moldy area can be accurately determined, and then based on the proportion, the mold level of the product can be quickly and accurately determined. Compared with the prior art, the present application can accurately determine the proportion of the moldy area, and then accurately and efficiently evaluate the mold level of the product, and has a wide range of applicability.

[0124] The device embodiments described above are merely illustrative, wherein the modules described as separate components may or may not be physically separated, i.e., may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the method of this embodiment.

[0125] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-monitorable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0126] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the scope of protection of the present application.

[0127] It is particularly pointed out that for those skilled in the art, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. A method for evaluating the moldiness level of a product, characterized in that: include: Acquire a moldy image of a product to be evaluated, and input the moldy image to be evaluated into a preset product area recognition model, so that the product area recognition model predicts the product area and product pixel information of the moldy image to be evaluated; wherein the product area recognition model is obtained by training based on a plurality of first product area images, and the plurality of first product area images are obtained by dividing moldy images of different products; Inputting the moldy image to be evaluated into a preset moldy area recognition model, so that the moldy area recognition model predicts the moldy area and moldy pixel information of the moldy image to be evaluated; wherein the moldy area recognition model is obtained by training a plurality of first moldy area images, and the plurality of first moldy area images are obtained by dividing the second product area images determined by the product area recognition model; Determine product pixels according to the product area and product pixel information of the moldy image to be evaluated, and determine moldy pixels according to the moldy area and moldy pixel information of the moldy image to be evaluated; Based on the product pixels and the moldy pixels, the proportion of the moldy area is determined, and based on the proportion, the mold level corresponding to the product is determined.

2. The method for evaluating the mold grade of a product according to claim 1, characterized in that: The method of enabling the moldy area recognition model to predict the moldy area and moldy pixel information of the moldy image to be evaluated is specifically: Determining first mask information corresponding to the moldy image to be evaluated based on the product area recognition model; A first feature vector is extracted from the moldy image to be evaluated, a product area is determined based on the first mask information and the first feature vector, and product pixel information in the product area is extracted, wherein the first feature vector includes a first color feature vector, a first texture feature vector, and a first shape feature vector.

3. The method for evaluating the mold grade of a product according to claim 1, characterized in that: The product area recognition model is obtained by training a plurality of first product area images, and the plurality of first product area images are obtained by dividing moldy images of different products, specifically: Acquire moldy images of different products, divide the moldy images into a first training set and a second training set, and determine a first product area image in the first training set; Training a first convolutional neural network based on the first product area image to obtain an initial product area recognition model; Inputting the second training set into the initial product area recognition model to obtain a third product area image, and mixing the third product area image with the first product area image to obtain a fourth product area image; The fourth product area image is interpolated to obtain a fifth product area image, product feature data of the fifth product area image is extracted, and the initial product area recognition model is trained based on the product feature data to obtain a product area recognition model.

4. The method for evaluating the mold grade of a product according to claim 1, characterized in that: The method of enabling the moldy area recognition model to predict the moldy area and moldy pixel information of the moldy image to be evaluated is specifically: Determining second mask information corresponding to the moldy image to be evaluated based on the moldy area recognition model; A second feature vector is extracted from the moldy image to be evaluated, a moldy area is determined based on the second mask information and the second feature vector, and moldy pixel information in the moldy area is extracted, wherein the second feature vector includes a second color feature vector, a second texture feature vector, and a second shape feature vector.

5. The method for evaluating the mold grade of a product according to claim 1, characterized in that: The moldy area recognition model is obtained by training a plurality of first moldy area images, and the plurality of first moldy area images are obtained by dividing the second product area images determined by the product area recognition model, specifically: Determine a second product area image based on the trained product area recognition model, divide the second product area image into a third training set and a fourth training set, and determine a first moldy area image in the third training set; Training a second convolutional neural network based on the first moldy area image to obtain an initial moldy area recognition model; Inputting the fourth training set into the initial moldy area recognition model to obtain a second moldy area image, and mixing the second moldy area image with the first moldy area image to obtain a third moldy area image; The third moldy area image is interpolated to obtain a fourth moldy area image, moldy feature data of the fourth moldy area image is extracted, and the initial moldy area recognition model is trained based on the moldy feature data to obtain a moldy area recognition model.

6. A product mold level assessment system, characterized in that: include: A first prediction module, a second prediction module, a determination module, and an evaluation module; The first prediction module is used to obtain a moldy image of a product to be evaluated, and input the moldy image to be evaluated into a preset product area recognition model, so that the product area recognition model predicts the product area and product pixel information of the moldy image to be evaluated; wherein the product area recognition model is obtained by training based on a plurality of first product area images, and the plurality of first product area images are obtained by dividing moldy images of different products; The second prediction module is used to input the moldy image to be evaluated into a preset moldy area recognition model, so that the moldy area recognition model predicts the moldy area and moldy pixel information of the moldy image to be evaluated; wherein the moldy area recognition model is obtained by training based on a plurality of first moldy area images, and the plurality of first moldy area images are obtained by dividing the second product area images determined by the product area recognition model; The determination module is used to determine product pixels according to the product area and product pixel information of the moldy image to be evaluated, and to determine moldy pixels according to the moldy area and moldy pixel information of the moldy image to be evaluated; The evaluation module is used to determine the proportion of the moldy area based on the product pixels and the moldy pixels, and determine the mold level corresponding to the product based on the proportion.

7. The product mold level assessment system according to claim 6, characterized in that: The first prediction module includes: a first determination unit and a first extraction unit; The first determining unit is used to determine first mask information corresponding to the moldy image to be evaluated based on the product area recognition model; The first extraction unit is used to extract a first feature vector from the moldy image to be evaluated, determine a product area based on the first mask information and the first feature vector, and extract product pixel information in the product area, wherein the first feature vector includes a first color feature vector, a first texture feature vector, and a first shape feature vector.

8. The product mold level assessment system according to claim 6, characterized in that: The first prediction module further includes: a first division unit, a first training unit, a first mixing unit, and a second training unit; The first division unit is used to obtain moldy images of different products, divide the moldy images into a first training set and a second training set, and determine a first product area image in the first training set; The first training unit is used to train a first convolutional neural network based on the first product area image to obtain an initial product area recognition model; The first mixing unit is used to input the second training set into the initial product area recognition model to obtain a third product area image, and mix the third product area image with the first product area image to obtain a fourth product area image; The second training unit is used to perform interpolation processing on the fourth product area image to obtain a fifth product area image, extract product feature data of the fifth product area image, and train the initial product area recognition model based on the product feature data to obtain a product area recognition model.

9. The product mold level evaluation system according to claim 6, characterized in that: The second prediction module includes: a second determination unit and a second extraction unit: The second determination unit is used to determine second mask information corresponding to the moldy image to be evaluated based on the moldy area recognition model; The second extraction unit is used to extract a second feature vector from the moldy image to be evaluated, determine a moldy area based on the second mask information and the second feature vector, and extract moldy pixel information in the moldy area, wherein the second feature vector includes a second color feature vector, a second texture feature vector, and a second shape feature vector.

10. The product mold level assessment system according to claim 6, characterized in that: The second prediction module further includes: a second division unit, a third training unit, a second mixing unit and a fourth training unit; The second division unit is used to determine a second product area image based on the trained product area recognition model, and divide the second product area image into a third training set and a fourth training set, and determine a first moldy area image in the third training set; The third training unit is used to train the second convolutional neural network based on the first moldy area image to obtain an initial moldy area recognition model; The second mixing unit is used to input the fourth training set into the initial moldy area recognition model to obtain a second moldy area image, and mix the second moldy area image with the first moldy area image to obtain a third moldy area image; The fourth training unit is used to perform interpolation processing on the third moldy area image to obtain a fourth moldy area image, extract moldy feature data of the fourth moldy area image, and train the initial moldy area recognition model based on the moldy feature data to obtain a moldy area recognition model.

Images