Product detection model global learning method and device, electronic equipment and storage medium

Abstract

The invention discloses a product detection model global learning method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring a product image training set of a target item; respectively extracting at least two-scale training guidance feature maps of each product positive sample image in the product image training set through a pre-trained AI guidance model; according to the training guidance feature map of each scale, determining a feature information fusion matrix and a guidance signal fusion matrix of a network layer corresponding to each scale in the AI precise model; and according to the feature information fusion matrix and the guidance signal fusion matrix, learning network parameters of a corresponding network layer in the AI precise-one model, and generating a precise-one product detection model. According to the scheme, the training speed of the precision-one product detection model is greatly improved, the problem of an inherent local optimal solution in an existing deep learning algorithm is solved, and it can be effectively guaranteed that the precision-one product detection model can learn a global optimal solution.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial product detection, and in particular to a global learning method, device, electronic device and storage medium for a product detection model. Background Art

[0002] In the field of intelligent detection of industrial products, the existing AI deep learning technology has a slow learning speed and generally only learns suboptimal local optimal solutions, which affects the efficiency of industrial production lines and the quality of products. Moreover, modern advanced intelligent flexible industrial production lines often need to replace new varieties or new models. However, the existing AI deep learning intelligent detection system requires a large amount of sample data to retrain the AI ​​model for new varieties or new types of data to adapt to new situations, and the waiting time is long, which greatly affects the detection quality and production efficiency of the production line and causes huge losses in economic benefits. Therefore, the detection system of this type of production line is in urgent need of high-quality intelligent solutions that can quickly adapt to new situations. Summary of the invention

[0003] The present invention provides a global learning method, device, electronic device and storage medium for a product inspection model, which not only greatly improves the training speed of the precise product inspection model, but also overcomes the problem of local optimal solutions inherent in existing deep learning algorithms, and can effectively ensure that the precise product inspection model can learn the global optimal solution.

[0004] According to one aspect of the present invention, a global learning method for a product detection model is provided, comprising:

[0005] Obtain a product image training set of a target project; wherein the product image training set includes at least two product positive sample images;

[0006] Extracting training guidance feature maps of at least two scales of each product positive sample image in the product image training set through a pre-trained AI guidance model; wherein the AI ​​guidance model is a deep learning model trained based on product images of different projects in the same field;

[0007] According to the training guidance feature graph of each scale, determining a feature information fusion matrix and a guidance signal fusion matrix of a network layer corresponding to each scale in an AI precision model; wherein the network structure of the AI ​​precision model is the same as the network structure of the AI ​​guidance model;

[0008] According to the feature information fusion matrix and the guidance signal fusion matrix, the network parameters of the corresponding network layer in the AI ​​precision model are learned to generate a precision product detection model.

[0009] According to another aspect of the present invention, a global learning device for a product detection model is provided, comprising:

[0010] A product image training set acquisition module, used to acquire a product image training set of a target project; wherein the product image training set contains at least two product positive sample images;

[0011] A training guidance feature map extraction module, used to extract training guidance feature maps of at least two scales of each product positive sample image in the product image training set through a pre-trained AI guidance model; wherein the AI ​​guidance model is a deep learning model trained based on product images of different projects in the same field;

[0012] An information fusion matrix determination module, used to determine the feature information fusion matrix and the guidance signal fusion matrix of the network layer corresponding to each scale in the AI ​​fine model according to the training guidance feature map of each scale; wherein the network structure of the AI ​​fine model is the same as the network structure of the AI ​​guidance model;

[0013] The refined product detection model learning module is used to learn the network parameters of the corresponding network layer in the AI ​​refined model according to the feature information fusion matrix and the guidance signal fusion matrix, and generate a refined product detection model.

[0014] According to another aspect of the present invention, there is provided an electronic device, the electronic device comprising:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the global learning method of the product detection model described in any embodiment of the present invention.

[0018] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the global learning method for product detection models described in any embodiment of the present invention when executed.

[0019] The global learning scheme of the product detection model of the embodiment of the present invention obtains a product image training set of the target project; wherein the product image training set contains at least two product positive sample images; the training guidance feature maps of at least two scales of each product positive sample image in the product image training set are extracted respectively by the pre-trained AI guidance model; wherein the AI ​​guidance model is a deep learning model based on product image training of different projects in the same field; according to the training guidance feature map of each scale, the feature information fusion matrix and the guidance signal fusion matrix of the network layer corresponding to each scale in the AI ​​precision model are determined; wherein the network structure of the AI ​​precision model is the same as the network structure of the AI ​​guidance model; according to the feature information fusion matrix and the guidance signal fusion matrix, the network parameters of the corresponding network layer in the AI ​​precision model are learned to generate a precision product detection model. The technical scheme provided by the embodiment of the present invention not only greatly improves the training speed of the precision product detection model, but also overcomes the problem of local optimal solution inherent in the existing deep learning algorithm, and can effectively ensure that the precision product detection model can learn the global optimal solution, thereby further improving the detection efficiency and detection quality of the product through the precision product detection model.

[0020] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 A flowchart of a global learning method for a product detection model provided in Embodiment 1 of the present invention;

[0023] Figure 2 A flowchart of a global learning method for a product detection model provided in Embodiment 2 of the present invention;

[0024] Figure 3 A schematic diagram of a global learning process of a product detection model provided by an embodiment of the present invention;

[0025] Figure 4 A flowchart of a global learning method for a product detection model and a product detection method provided in Embodiment 3 of the present invention;

[0026] Figure 5A schematic diagram of a product detection process provided by an embodiment of the present invention;

[0027] Figure 6 A schematic diagram of the structure of an AI intelligent detection system for industrial product detection provided in Embodiment 4 of the present invention;

[0028] Figure 7 A schematic diagram of the structure of a global learning device for a product detection model provided in Embodiment 5 of the present invention;

[0029] Figure 8 A schematic diagram of the structure of an electronic device for implementing the global learning method of a product detection model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0031] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0032] Embodiment 1

[0033] Figure 1 A flowchart of a method for global learning of a product detection model is provided for the first embodiment of the present invention. This embodiment is applicable to the case of generating a product detection model. The method can be performed by a global learning device for a product detection model. The global learning device for a product detection model can be implemented in the form of hardware and / or software. The global learning device for a product detection model can be configured in an electronic device. Figure 1 As shown, the method includes:

[0034] S110, obtaining a product image training set of a target project; wherein the product image training set includes at least two product positive sample images.

[0035] In an embodiment of the present invention, industrial production involves multiple fields, and each field contains different projects. For example, in the technical field of mobile terminals (such as mobile phones), multiple different projects are involved, such as the same project corresponds to the same model of mobile phones. At least two product positive sample images of the target project are obtained, and a product image training set is generated based on at least two product positive sample images. Among them, the product positive sample images included in the product image training set are representative and diverse. Exemplarily, the product positive sample images included in the product image training set are sample images of positive products (that is, products without quality problems or non-abnormal products) of the target project taken by the camera at different shooting angles and different lighting conditions.

[0036] S120. Extracting training guidance feature maps of at least two scales for each product positive sample image in the product image training set through a pre-trained AI guidance model; wherein the AI ​​guidance model is a deep learning model trained based on product images of different projects in the same field.

[0037] In an embodiment of the present invention, a pre-trained AI guidance model is obtained, wherein the pre-trained AI guidance model can be understood as an already trained AI guidance model, and the AI ​​guidance model is a deep learning model trained based on product images of different projects in the same field. Exemplarily, based on at least two product images of different projects in the same field (such as multiple different models of mobile phones), the AI ​​guidance model is trained to generate an AI guidance model. Among them, the product images of different projects in the same field used to train the AI ​​guidance model include product positive sample images and product negative sample images. Each product positive sample image in the product image training set of the target project is input into the AI ​​guidance model respectively, and the training guidance feature map of at least two scales corresponding to each product positive sample image is obtained through the AI ​​guidance model. Exemplarily, the training guidance feature map of three scales corresponding to each product positive sample image is obtained through the AI ​​guidance model, such as obtaining a 128*128 training guidance feature map, a 64*64 training guidance feature map, and a 32*32 training guidance feature map. For example, if the product image training set contains 100 positive sample images of products, then the AI ​​guidance model can obtain 100*C1 128*128 training guidance feature maps, 100*C2 64*64 training guidance feature maps, and 100*C3 32*32 training guidance feature maps, where Ci is the number of channels of the training guidance feature map at the corresponding scale. It should be noted that the training guidance feature map of each scale of each product positive sample image extracted by the AI ​​guidance model is a multi-channel training guidance feature map, that is, the training guidance feature map of multiple channels of each scale of each product positive sample image extracted by the AI ​​guidance model.

[0038] S130. Determine, according to the training guidance feature graph of each scale, a feature information fusion matrix and a guidance signal fusion matrix of the network layer corresponding to each scale in the AI ​​precision model; wherein the network structure of the AI ​​precision model is the same as the network structure of the AI ​​guidance model.

[0039] In an embodiment of the present invention, an AI precision model is obtained, wherein the AI ​​precision model is an untrained deep learning model, and the network structure of the AI ​​precision model is the same as the network structure of the AI ​​guidance model. Exemplarily, the AI ​​guidance model includes three network layers, and the three network layers are a 128*128 network layer, a 64*64 network layer, and a 32*32 network layer, and the 128*128 network layer, the 64*64 network layer, and the 32*32 network layer are sequentially connected from input to output, then the AI ​​precision model is also composed of a 128*128 network layer, a 64*64 network layer, and a 32*32 network layer sequentially connected from input to output. Among them, the initial network parameters in the AI ​​precision model can be any numerical value, and the embodiment of the present invention does not limit the initial network parameters in the AI ​​precision model.

[0040] For each scale of the training guidance feature graphs of at least two scales, the feature information fusion matrix A and the guidance information fusion matrix B of the network layer corresponding to each scale in the AI ​​precision model are determined according to the training guidance feature graphs of the scale. It can be understood that all training guidance feature graphs of the same scale are analyzed to determine the feature information fusion matrix and the guidance signal fusion matrix of the network layer corresponding to each scale in the AI ​​precision model. Exemplarily, according to 100*C1 sheets of 128*128 training guidance feature graphs, the feature information fusion matrix A1 and the guidance signal fusion matrix B1 of the network layer corresponding to the scale of 128*128 in the AI ​​precision model are determined; according to 100*C2 sheets of 64*64 training guidance feature graphs, the feature information fusion matrix A2 and the guidance signal fusion matrix B2 of the network layer corresponding to the scale of 64*64 in the AI ​​precision model are determined; according to 100*C3 sheets of 32*32 training guidance feature graphs, the feature information fusion matrix A3 and the guidance signal fusion matrix B3 of the network layer corresponding to the scale of 32*32 in the AI ​​precision model are determined. Among them, the number of feature information fusion matrices and the number of guidance signal fusion matrices are the same as the number of network layers included in the AI ​​precision model, and are also the same as the number of scales involved in the training guidance feature map.

[0041] S140. According to the feature information fusion matrix and the guidance signal fusion matrix, the network parameters of the corresponding network layer in the AI ​​precision model are learned to generate a precision product detection model.

[0042] In an embodiment of the present invention, the characteristic information fusion matrix and the guidance signal fusion matrix of the network layer of each scale in the AI ​​fine model are processed according to the preset processing strategy, and the network parameters of the corresponding network layer in the AI ​​fine model are learned. Exemplarily, according to the characteristic information fusion matrix A1 and the guidance signal fusion matrix B1 of the network layer corresponding to the scale of 128*128 in the AI ​​fine model, the network parameters of the network layer corresponding to the scale of 128*128 in the AI ​​fine model are calculated; according to the characteristic information fusion matrix A2 and the guidance signal fusion matrix B2 of the network layer corresponding to the scale of 64*64 in the AI ​​fine model, the network parameters of the network layer corresponding to the scale of 64*64 in the AI ​​fine model are calculated; according to the characteristic information fusion matrix A3 and the guidance signal fusion matrix B3 of the network layer corresponding to the scale of 32*32 in the AI ​​fine model, the network parameters of the network layer corresponding to the scale of 32*32 in the AI ​​fine model are calculated. According to the above method, the network parameters of each network layer in the AI ​​fine model can be quickly calculated, so that the AI ​​fine model that determines the network parameters of each network layer is used as the fine product detection model. It can be understood that the precise product detection model is a trained AI precise model, that is, the precise product detection model is an AI precise model that can be used for product detection.

[0043] The global learning method of the product detection model of the embodiment of the present invention obtains a product image training set of the target project; wherein the product image training set contains at least two product positive sample images; the training guidance feature maps of at least two scales of each product positive sample image in the product image training set are extracted respectively by the pre-trained AI guidance model; wherein the AI ​​guidance model is a deep learning model based on product image training of different projects in the same field; according to the training guidance feature map of each scale, the feature information fusion matrix and the guidance signal fusion matrix of the network layer corresponding to each scale in the AI ​​precision model are determined; wherein the network structure of the AI ​​precision model is the same as the network structure of the AI ​​guidance model; according to the feature information fusion matrix and the guidance signal fusion matrix, the network parameters of the corresponding network layer in the AI ​​precision model are learned to generate a precision product detection model. The technical solution provided by the embodiment of the present invention not only greatly improves the training speed of the precision product detection model, but also overcomes the problem of the local optimal solution inherent in the existing deep learning algorithm, and can effectively ensure that the precision product detection model can learn the global optimal solution, thereby further improving the detection efficiency and detection quality of the product through the precision product detection model.

[0044] Embodiment 2

[0045] Figure 2 A flowchart of a global learning method for a product detection model provided in the second embodiment of the present invention is shown in FIG. Figure 2As shown, the method includes:

[0046] S210, obtaining a product image training set of a target project; wherein the product image training set includes at least two product positive sample images.

[0047] S220, traversing each product positive sample image in the product image training set, and extracting the training guidance feature map of the product positive sample image with the same scale as the corresponding network layer through the network layer of each scale in the pre-trained AI guidance model.

[0048] In an embodiment of the present invention, the AI ​​guidance model is composed of at least two network layers of different scales, wherein the scales of the network layers are different and the sizes of the output feature maps are different. Each positive sample image of a product in the product image training set is traversed, each positive sample image of a product is input into the AI ​​guidance model, and the training guidance feature map of the positive sample image of the product is extracted according to the network layer of each scale in the AI ​​guidance model. Exemplarily, Figure 3 A schematic diagram of a global learning process of a product detection model provided by an embodiment of the present invention is shown in FIG. Figure 3 As shown, the AI ​​guidance model G contains a total of m network layers, namely the first network layer G1, the second network layer G2, ..., the mth network layer Gm, wherein each network layer in the AI ​​guidance model G can be composed of a convolutional layer, or a convolutional layer and an activation function layer, or a convolutional layer, an activation function layer and a pooling layer. The m network layers in the AI ​​guidance model G are connected sequentially from input to output, that is, the output end of the first network layer G1 is connected to the input end of the second network layer G2, the output end of the second network layer G2 is connected to the input end of the third network layer G3, ..., the output end of the m-1th network layer G(m-1) is connected to the input end of the m-th network layer Gm. The product positive sample image is input into the first network layer G1 in the AI ​​guidance model G, and the training guidance feature map T1 with the same scale as the first network layer G1 is extracted from the product positive sample image through the first network layer G1; the training guidance feature map T1 output by the first network layer G1 is input into the second network layer G2, and the training guidance feature map T2 with the same scale as the second network layer G2 is extracted from the training guidance feature map T1 through the second network layer G2; and so on, the training guidance feature map Tm with the same scale as the mth network layer Gm can be obtained through the mth network layer Gm. The scale of the network layer is used to characterize the size of the feature map output by the network layer. For example, if the scale of a network layer is 64*64, then the network layer outputs a 64*64 training guidance feature map.

[0049] S230. Determine structural information of each network layer in the AI ​​refined model; wherein the network structure of the AI ​​refined model is the same as the network structure of the AI ​​guidance model.

[0050] In an embodiment of the present invention, the structural information of each network layer in the AI ​​precision model is determined, wherein the structural information includes the structural parameters and the number of channels of the convolutional layer of each network layer. The structural information of each network layer is used to characterize the structural parameters of each network layer that need to be determined. Figure 3 As shown, the AI ​​precise model S contains a total of m network layers, namely the first network layer S1, the second network layer S2, ..., the mth network layer Sm, and the structural information H1 of the first network layer S1, the structural information H2 of the second network layer S2, ..., and the structural information Hm of the mth network layer Sm are determined respectively. The m network layers in the AI ​​precise model S are connected sequentially from input to output, that is, the output end of the first network layer S1 is connected to the input end of the second network layer S2, the output end of the second network layer S2 is connected to the input end of the third network layer S3, ..., the output end of the m-1th network layer S(m-1) is connected to the input end of the m-1th network layer Sm. Optionally, the structural information of each network layer in the AI ​​precise model can be determined according to the information input by the user, that is, the structural information of each network layer in the AI ​​precise model can be the network structure information pre-set by the user.

[0051] S240. Determine a feature information fusion matrix and a guidance signal fusion matrix of the network layer corresponding to each scale in the AI ​​fine model according to the training guidance feature map of each scale and the structural information of the network layer of the corresponding scale in the AI ​​fine model.

[0052] For example, Figure 3As shown, the global learning module determines the feature information fusion matrix A1 and the guidance signal fusion matrix B1 corresponding to the first network layer S1 in the AI ​​fine model according to all the training guidance feature graphs T1 corresponding to the product image training set output by the first network layer G1 of the AI ​​guidance model and the structural information of the first network layer S1 in the AI ​​fine model; the global learning module determines the feature information fusion matrix A2 and the guidance signal fusion matrix B2 corresponding to the second network layer S2 in the AI ​​fine model according to all the training guidance feature graphs T2 corresponding to the product image training set output by the second network layer G2 of the AI ​​guidance model and the structural information of the second network layer S2 in the AI ​​fine model; and by analogy, the global learning module determines the feature information fusion matrix Am and the guidance signal fusion matrix Bm corresponding to the mth network layer Sm in the AI ​​fine model according to all the training guidance feature graphs Tm corresponding to the product image training set output by the mth network layer Gm of the AI ​​guidance model and the structural information of the mth network layer Sm in the AI ​​fine model. Among them, the structural information of each network layer in the AI ​​fine model is used to characterize the size of the feature information fusion matrix and the guidance signal fusion matrix of the corresponding network layer in the AI ​​fine model.

[0053] S250. According to the feature information fusion matrix, the guidance signal fusion matrix and the structural information of the network layer of the corresponding scale in the AI ​​refined model, learn the network parameters of the corresponding network layer in the AI ​​refined model to generate a refined product detection model.

[0054] In the embodiment of the present invention, the structural information of each network layer in the AI ​​precise model is used to characterize the structural parameters of each network layer that needs to be determined for the corresponding network layer. Therefore, the processing strategy for the characteristic information fusion matrix and the guidance signal fusion matrix of the corresponding network layer in the AI ​​precise model can be determined according to the structural information of each network layer in the AI ​​precise model, and the characteristic information fusion matrix and the guidance signal fusion matrix of the corresponding network layer in the AI ​​precise model are processed according to the corresponding processing strategy, so as to calculate the network parameters of the corresponding network layer in the AI ​​precise model, thereby realizing the global training and learning of the network parameters of each network layer in the AI ​​precise model. Figure 3As shown, the global learning module determines the feature information fusion matrix A1 and the guidance signal fusion matrix B1 according to the structural information H1 of the first network layer S1 in the AI ​​precise model and the guidance feature atlas T1 corresponding to the first network layer G1 of the AI ​​guidance model, and determines the various network parameters of the first network layer S1 in the AI ​​precise model according to the feature information fusion matrix A1, the guidance signal fusion matrix B1 and the structural information H1 of the first network layer S1 in the AI ​​precise model; the global learning module determines the feature information fusion matrix A2 and the guidance signal fusion matrix B2 according to the structural information H2 of the second network layer S2 in the AI ​​precise model and the guidance feature atlas T2 corresponding to the second network layer G2 of the AI ​​guidance model. , and determine the various network parameters of the second network layer S2 in the AI ​​precise model according to the feature information fusion matrix A2 and the guidance signal fusion matrix B2 and the structural information H2 of the second network layer S2 in the AI ​​precise model; and by analogy, the global learning module determines the feature information fusion matrix Am and the guidance signal fusion matrix Bm according to the structural information Hm of the m-th network layer Sm in the AI ​​precise model and the guidance feature atlas Tm corresponding to the m-th network layer Gm of the AI ​​guidance model, and determines the various network parameters of the m-th network layer Sm in the AI ​​precise model according to the feature information fusion matrix Am and the guidance signal fusion matrix Bm and the structural information Hm of the m-th network layer Sm in the AI ​​precise model.

[0055] The global learning method for product inspection models provided by the embodiment of the present invention not only greatly improves the training speed of the precise product inspection model, but also overcomes the problem of local optimal solutions inherent in existing deep learning algorithms. It can effectively ensure that the precise product inspection model can learn the global optimal solution, thereby quickly realizing the rapid launch or seamless connection of the precise product inspection AI model when new inspection projects are launched or product models are updated for existing projects, significantly reducing the waiting time of existing deep learning technologies and improving economic benefits.

[0056] Embodiment 3

[0057] Figure 4 A flowchart of a global learning method for a product detection model and a product detection method provided in Embodiment 3 of the present invention is shown in FIG. Figure 4 As shown, the method includes:

[0058] S410, obtaining a product image training set of a target project; wherein the product image training set includes at least two product positive sample images.

[0059] S420. Extracting training guidance feature maps of at least two scales for each product positive sample image in the product image training set through a pre-trained AI guidance model; wherein the AI ​​guidance model is a deep learning model trained based on product images of different projects in the same field.

[0060] S430. Determine, according to the training guidance feature graph of each scale, a feature information fusion matrix and a guidance signal fusion matrix of the network layer corresponding to each scale in the AI ​​precision model; wherein the network structure of the AI ​​precision model is the same as the network structure of the AI ​​guidance model.

[0061] S440. According to the feature information fusion matrix and the guidance signal fusion matrix, the network parameters of the corresponding network layer in the AI ​​precision model are learned to generate a precision product detection model.

[0062] S450: Acquire an image of a product to be inspected.

[0063] In the embodiment of the present invention, an image of the product to be detected is obtained, and the image of the product to be detected is used as the image of the product to be detected. The image of the product to be detected may be one or more images, and the embodiment of the present invention does not limit the number of images of the product to be detected.

[0064] S460. Extract detection guidance feature maps of at least two scales of the product image to be detected through the AI ​​guidance model, and extract target feature maps of at least two scales of the product image to be detected through the precise product detection model.

[0065] In an embodiment of the present invention, the image of the product to be detected is input into the AI ​​guidance model, and the detection guidance feature map of the product image to be detected is extracted according to each scale of the network layer in the AI ​​guidance model. The image of the product to be detected is input into the precision product detection model, and the target feature map of the product image to be detected is extracted according to each scale of the network layer in the precision product detection model, wherein the size of the target feature map of the product image to be detected extracted by the network layer of each scale is the same as the size of the detection guidance feature map of the corresponding network layer. Since the network structure of the AI ​​guidance model is the same as that of the precision product detection model, the detection guidance feature map extracted by each network layer of the AI ​​guidance model has the same scale as the target feature map extracted by the corresponding network layer of the precision product detection model.

[0066] S470: Determine a detection result of the product to be detected corresponding to the product image to be detected according to the detection guidance feature map and the target feature map.

[0067] For example, each target feature map of at least two scales may be compared with a detection guidance feature map of a corresponding scale in the detection guidance feature map of at least two scales. Figure 1One is to make a comparison, and determine the detection result of the product to be detected corresponding to the product image to be detected according to the comparison result. Among them, the detection result of the product to be detected corresponding to the product image to be detected includes qualified detection and unqualified detection, or the product to be detected is genuine and the product to be detected is defective. Optionally, the detection result of the product to be detected corresponding to the product image to be detected is determined according to the detection guidance feature map and the target feature map, including: fusing the detection guidance feature map of at least two scales and the target feature map of at least two scales to generate a detection heat map; based on the detection heat map, determining the detection result of the product to be detected corresponding to the product image to be detected.

[0068] Optionally, the detection guidance feature maps of the at least two scales and the target feature maps of the at least two scales are fused to generate a detection heat map, including: comparing the detection guidance feature map of each scale in the detection guidance feature map of the at least two scales with the target feature map of the corresponding scale in the target feature map of the at least two scales, and generating a corresponding initial heat map according to the comparison result; and fusing the initial heat maps corresponding to all scales to generate a detection heat map. Exemplarily, the detection guidance feature map of the corresponding scale determined by the first network layer G1 of the AI ​​guidance model is compared with the target feature map of the corresponding scale determined by the first network layer S1 of the precision product detection model, and the initial thermal map P1 is generated according to the comparison result; the detection guidance feature map of the corresponding scale determined by the second network layer G2 of the AI ​​guidance model is compared with the target feature map of the corresponding scale determined by the second network layer S2 of the precision product detection model, and the initial thermal map P2 is generated according to the comparison result; and so on, the detection guidance feature map of the corresponding scale determined by the mth network layer Gm of the AI ​​guidance model is compared with the target feature map of the corresponding scale determined by the mth network layer Sm of the precision product detection model, and the initial thermal map Pm is generated according to the comparison result. It should be noted that the greater the difference between the detection guidance feature map and the target feature map, the darker the overall color of the generated initial thermal map; conversely, the smaller the difference between the detection guidance feature map and the target feature map, the lighter the overall color of the generated initial thermal map. The initial heat maps of all scales are fused, that is, the initial heat map P1, the initial heat map P2, ..., the initial heat map Pm are fused to generate a detection heat map. Then, according to the pre-set detection index, the detection heat map is analyzed to determine the detection result of the product to be detected corresponding to the product image to be detected. Exemplarily, the hot area in the detection heat map whose color depth is greater than the preset color depth threshold is judged, and the ratio of the area of ​​the hot area to the area of ​​the entire detection heat map is calculated. If the ratio is greater than the preset threshold, it can be determined that the product to be detected corresponding to the product image to be detected is a defective product; if the ratio is less than the preset threshold, it can be determined that the product to be detected corresponding to the product image to be detected is a genuine product.

[0069] Optionally, the detection guidance feature maps of the at least two scales and the target feature maps of the at least two scales are fused to generate a detection heat map, including: fusing the detection guidance feature maps of the at least two scales to generate a fused guidance feature map; fusing the target feature maps of the at least two scales to generate a fused target feature map; comparing the fused guidance feature map and the fused target feature map, and generating a detection heat map according to the comparison result. Exemplarily, the detection guidance feature map of the corresponding scale determined by the first network layer G1 of the AI ​​guidance model, the detection guidance feature map of the corresponding scale determined by the second network layer G2 of the AI ​​guidance model, ..., and the detection guidance feature map of the corresponding scale determined by the mth network layer Gm of the AI ​​guidance model are fused according to a preset fusion strategy to generate a fused guidance feature map. Similarly, according to the preset fusion strategy, the target feature map of the corresponding scale determined by the first network layer S1 of the precise product detection model, the target feature map of the corresponding scale determined by the second network layer S2 of the precise product detection model, ..., the target feature map of the corresponding scale determined by the mth network layer Sm of the precise product detection model are fused to generate a fused target feature map. The fused target feature map is compared with the fused guidance feature map, and a detection heat map is generated according to the comparison result. It should be noted that the greater the difference between the fused target feature map and the fused guidance feature map, the darker the overall color of the generated detection heat map; conversely, the smaller the difference between the fused target feature map and the fused guidance feature map, the lighter the overall color of the generated detection heat map. Then, according to the pre-set detection index, the detection heat map is analyzed to determine the detection result of the product to be detected corresponding to the product image to be detected.

[0070] For example, Figure 5 A schematic diagram of a product detection process provided by an embodiment of the present invention. Figure 5 Please understand and I will not elaborate on it again.

[0071] Through the technical solution provided by the embodiment of the present invention, the detection result of the product to be detected can be determined quickly and accurately through the learned precise product detection model.

[0072] Embodiment 4

[0073] Figure 6 This is a schematic diagram of the structure of an AI intelligent detection system for industrial product detection provided by the fourth embodiment of the present invention. Figure 6As shown, the AI ​​intelligent detection system includes: an AI guidance model unit, an AI guidance model learning unit, an AI precision model unit, a multi-scale feature map extraction unit, a fast global learning unit, a sample database management unit, a project training image acquisition unit, a project training image selection unit, a domain database unit, and a quality and accuracy control unit.

[0074] AI guidance model unit: used to store the trained AI guidance model, through which the guidance feature map can be extracted. For example, in the process of training the AI ​​precision model, the guidance feature map of each product positive sample image contained in the product image training set can be extracted through the AI ​​guidance model to guide the training of the AI ​​precision model according to the guidance feature map. For another example, when inspecting industrial products, the detection guidance feature map of the image of the product to be inspected can be extracted through the AI ​​guidance model to inspect the product to be inspected according to the detection guidance feature map and the target feature map.

[0075] AI guidance model learning unit: use samples in the domain database unit to learn the untrained AI guidance model, generate a trained AI guidance model, and send the trained AI guidance model to the AI ​​guidance model unit. Since it is generally difficult to obtain a large number of samples in the specific project of abnormal product detection on industrial production lines, the learning process and online process of AI intelligent detection are greatly hindered and delayed. In an embodiment of the present invention, a large amount of sample data can be accumulated by collecting samples of many similar but different projects in the same industry. Then, based on a large amount of widely representative sample data in the same field, a deep convolutional neural network learning algorithm is used to learn and train the AI ​​guidance model, so that it can essentially mine, extract and learn the common features of different project data samples and other field knowledge. Therefore, the AI ​​guidance model not only overcomes the problem of insufficient samples in the specific project of industrial product detection, but also improves the quality and generalization ability of the AI ​​model.

[0076] AI Precision Model Unit: used to store AI Precision Model, where AI Precision Model is a modular AI deep learning system architecture with the same structure as the AI ​​guidance model in terms of deep learning network structure. Each AI Precision Model is only proficient in anomaly detection of one type of industrial product or anomaly detection of products of the same project to improve detection accuracy and efficiency. It is understandable that AI Precision Model is a specific project for industrial production line detection, and the corresponding network structure matching the project is selected according to user needs.

[0077] Multi-scale feature map extraction unit: used to extract training guidance feature maps of at least two scales of the product image training set through the AI ​​guidance model, and send the training guidance feature maps to the fast global learning unit.

[0078] Fast global learning unit: Based on the training guidance feature maps of at least two scales sent by the multi-scale feature map extraction unit, the network parameters of each network layer of the AI ​​precision model are globally learned, thereby determining the network parameters of each network layer in the AI ​​precision model, generating a precision product detection model, and sending the precision product detection model to the AI ​​precision model unit for storage. The fast global training method provided by the embodiment of the present invention can significantly reduce the learning time of the precision product detection model, enhance the generalization ability of the precision product detection model, and overcome the problem that the product detection model falls into the local optimal solution in the existing deep learning process.

[0079] Sample database management unit: responsible for the collection, management and optimization of sample database, ensuring the quality and consistency of data samples.

[0080] Project training image acquisition unit: used to collect sample images of new projects and pre-process the collected sample images of new projects to improve data consistency and sample data quality.

[0081] Project training image selection unit: used to select appropriate sample images from the sample image set collected by the project training image collection unit according to the specific situation of the project, as the product image training set for training the refined product detection model.

[0082] Domain database unit: stores sample images of different items in the domain.

[0083] Quality and accuracy control unit: used to control and set the configuration parameters of the entire AI intelligent detection system, including sample quality control parameters, real-time detection accuracy settings, learning process accuracy settings, and learning hyperparameter settings. The configuration parameters are set with default configuration values, which are generally not required to be modified by the user. The user can also flexibly configure learning hyperparameters and other configuration parameters according to the specific needs of the project to effectively improve the learning and training efficiency of the Jingyi product detection model.

[0084] The AI ​​intelligent detection system provided by the embodiment of the present invention can quickly estimate the training and learning time of the refined product detection model under the given number of training samples and deep neural network structure, greatly improving the management level of production line launch and changeover planning. In addition, in the process of training the refined product detection model, since there is no training convergence problem, the global optimal solution can be found in one training. Therefore, there is no need for manual intervention to check whether the training converges, whether it converges to the correct solution, etc., and there is no need to train the neural network model multiple times and select the best training and learning results from it, thereby not only greatly improving the training speed of the refined product detection model, but also overcoming the problem of local optimal solutions inherent in existing deep learning algorithms, and effectively ensuring that the refined product detection model can learn the global optimal solution, so that the refined product detection model can be quickly launched or seamlessly connected when new detection projects or existing projects update product models, significantly reducing the waiting time of the current existing deep learning technology and improving economic benefits. In addition, when training and learning the Jingyi product detection model, each sample data only needs to be learned once to remember the complete information of the sample data. There is no need to learn multiple times, only new sample data needs to be learned. Therefore, there is no need to save and learn old sample data additionally, which reduces the maintenance cost of sample data and greatly saves the storage space and cost of training samples.

[0085] Embodiment 5

[0086] Figure 7 This is a schematic diagram of the structure of a global learning device for a product detection model provided in Embodiment 5 of the present invention. Figure 7 As shown, the device comprises:

[0087] A product image training set acquisition module 710 is used to acquire a product image training set of a target project; wherein the product image training set includes at least two product positive sample images;

[0088] A training guidance feature map extraction module 720 is used to extract training guidance feature maps of at least two scales of each product positive sample image in the product image training set through a pre-trained AI guidance model; wherein the AI ​​guidance model is a deep learning model trained based on product images of different projects in the same field;

[0089] An information fusion matrix determination module 730 is used to determine, according to the training guidance feature graph of each scale, a feature information fusion matrix and a guidance signal fusion matrix of a network layer corresponding to each scale in the AI ​​fine model; wherein the network structure of the AI ​​fine model is the same as the network structure of the AI ​​guidance model;

[0090] The refined product detection model learning module 740 is used to learn the network parameters of the corresponding network layer in the AI ​​refined model according to the feature information fusion matrix and the guidance signal fusion matrix, and generate a refined product detection model.

[0091] Optionally, the AI ​​guidance model includes at least two network layers of different scales;

[0092] The training guidance feature map extraction module is used to:

[0093] Each product positive sample image in the product image training set is traversed, and a training guidance feature map of the product positive sample image having the same scale as the corresponding network layer is extracted through the network layer of each scale in the pre-trained AI guidance model.

[0094] Optionally, the device further comprises:

[0095] a structure information determination module, configured to determine the structure information of each network layer in the AI ​​fine model before learning the network parameters of the corresponding network layer in the AI ​​fine model according to the feature information fusion matrix and the guidance signal fusion matrix;

[0096] The information fusion matrix determination module is used to:

[0097] Determine a feature information fusion matrix and a guidance signal fusion matrix of the network layer corresponding to each scale in the AI ​​fine-tuned model according to the training guidance feature map of each scale and the structural information of the network layer of the corresponding scale in the AI ​​fine-tuned model;

[0098] The refined product detection model learning module is used to:

[0099] According to the feature information fusion matrix, the guidance signal fusion matrix and the structural information of the network layer of the corresponding scale in the AI ​​refined model, the network parameters of the corresponding network layer in the AI ​​refined model are learned to generate a refined product detection model.

[0100] Optionally, the device further comprises:

[0101] The module for acquiring the image of the product to be inspected is used to acquire the image of the product to be inspected after the precise product inspection model is generated;

[0102] A target feature map extraction module, used to extract detection guidance feature maps of at least two scales of the product image to be detected through the AI ​​guidance model, and to extract target feature maps of at least two scales of the product image to be detected through the precise product detection model;

[0103] The detection result determination module is used to determine the detection result of the product to be detected corresponding to the image of the product to be detected according to the detection guidance feature map and the target feature map.

[0104] Optionally, the detection result determination module includes:

[0105] A detection heat map generating unit, configured to fuse the detection guidance feature maps of the at least two scales and the target feature maps of the at least two scales to generate a detection heat map;

[0106] A detection result determination unit is used to determine a detection result of the product to be detected corresponding to the image of the product to be detected based on the detection heat map.

[0107] Optionally, the detection heat map generating unit is used to:

[0108] Comparing the detection guidance feature map of each scale in the at least two scales of the detection guidance feature map with the target feature map of a corresponding scale in the at least two scales of the target feature map, and generating a corresponding initial heat map according to the comparison result;

[0109] The initial heat maps corresponding to all scales are fused to generate a detection heat map.

[0110] Optionally, the detection heat map generating unit is used to:

[0111] Fusing the detection guidance feature maps at the at least two scales to generate a fused guidance feature map;

[0112] Fusing the target feature maps at the at least two scales to generate a fused target feature map;

[0113] The fusion guidance feature map and the fusion target feature map are compared, and a detection heat map is generated according to the comparison result.

[0114] The global learning device for product detection models provided in the embodiment of the present invention can execute the global learning method for product detection models provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0115] Embodiment 6

[0116] Figure 8A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0117] like Figure 8 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0118] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0119] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running deep learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The processor 11 executes the various methods and processes described above, such as a global learning method for a product detection model.

[0120] In some embodiments, the global learning method of the product detection model may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the global learning method of the product detection model described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the global learning method of the product detection model in any other appropriate manner (e.g., by means of firmware).

[0121] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0122] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0123] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0124] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0125] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0126] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0127] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0128] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A global learning method for a product detection model, characterized in that: include: Obtain a product image training set of a target project; wherein the product image training set includes at least two product positive sample images; Extracting training guidance feature maps of at least two scales of each product positive sample image in the product image training set respectively through a pre-trained AI guidance model; wherein the AI ​​guidance model is a deep learning model trained based on product images of different projects in the same field; According to the training guidance feature graph of each scale, determining a feature information fusion matrix and a guidance signal fusion matrix of a network layer corresponding to each scale in an AI precision model; wherein the network structure of the AI ​​precision model is the same as the network structure of the AI ​​guidance model; According to the feature information fusion matrix and the guidance signal fusion matrix, the network parameters of the corresponding network layer in the AI ​​precision model are learned to generate a precision product detection model.

2. The method according to claim 1, characterized in that The AI ​​guidance model includes at least two network layers of different scales; The pre-trained AI guidance model is used to extract training guidance feature maps of at least two scales for each product positive sample image in the product image training set, including: Each product positive sample image in the product image training set is traversed, and a training guidance feature map of the product positive sample image having the same scale as the corresponding network layer is extracted through the network layer of each scale in the pre-trained AI guidance model.

3. The method according to claim 1, characterized in that Before determining the feature information fusion matrix and the guidance signal fusion matrix of the network layer corresponding to each scale in the AI ​​precision model according to the training guidance feature map of each scale, the method further includes: Determine the structural information of each network layer in the AI ​​precision model; According to the training guidance feature map of each scale, determining the feature information fusion matrix and the guidance signal fusion matrix of the network layer corresponding to each scale in the AI ​​precision model, including: Determine a feature information fusion matrix and a guidance signal fusion matrix of the network layer corresponding to each scale in the AI ​​fine model according to the training guidance feature map of each scale and the structural information of the network layer of the corresponding scale in the AI ​​fine model; According to the feature information fusion matrix and the guidance signal fusion matrix, learning the network parameters of the network layer corresponding to the AI ​​refined model to generate a refined product detection model, including: According to the feature information fusion matrix, the guidance signal fusion matrix and the structural information of the network layer of the corresponding scale in the AI ​​refined model, the network parameters of the corresponding network layer in the AI ​​refined model are learned to generate a refined product detection model.

4. The method according to claim 1, characterized in that After generating a refined product detection model, it also includes: Acquire the image of the product to be inspected; Extracting detection guidance feature maps of at least two scales of the product image to be detected by the AI ​​guidance model, and extracting target feature maps of at least two scales of the product image to be detected by the precise product detection model; A detection result of the product to be detected corresponding to the product image to be detected is determined according to the detection guidance feature map and the target feature map.

5. The method according to claim 4, characterized in that Determining a detection result of the product to be detected corresponding to the product image to be detected according to the detection guidance feature map and the target feature map includes: Fusion the detection guidance feature maps of the at least two scales and the target feature maps of the at least two scales to generate a detection heat map; A detection result of the product to be detected corresponding to the product image to be detected is determined based on the detection heat map.

6. The method according to claim 5, characterized in that The detection guidance feature maps of the at least two scales and the target feature maps of the at least two scales are fused to generate a detection heat map, including: Comparing the detection guidance feature map of each scale in the at least two scales of the detection guidance feature map with the target feature map of a corresponding scale in the at least two scales of the target feature map, and generating a corresponding initial heat map according to the comparison result; The initial heat maps corresponding to all scales are fused to generate a detection heat map.

7. The method according to claim 5, characterized in that The detection guidance feature maps of the at least two scales and the target feature maps of the at least two scales are fused to generate a detection heat map, including: Fusing the detection guidance feature maps at the at least two scales to generate a fused guidance feature map; Fusing the target feature maps at the at least two scales to generate a fused target feature map; The fusion guidance feature map and the fusion target feature map are compared, and a detection heat map is generated according to the comparison result.

8. A global learning device for a product inspection model, characterized in that: include: A product image training set acquisition module, used to acquire a product image training set of a target project; wherein the product image training set contains at least two product positive sample images; A training guidance feature map extraction module, used to extract training guidance feature maps of at least two scales of each product positive sample image in the product image training set through a pre-trained AI guidance model; wherein the AI ​​guidance model is a deep learning model trained based on product images of different projects in the same field; An information fusion matrix determination module, used to determine the feature information fusion matrix and the guidance signal fusion matrix of the network layer corresponding to each scale in the AI ​​fine model according to the training guidance feature map of each scale; wherein the network structure of the AI ​​fine model is the same as the network structure of the AI ​​guidance model; The refined product detection model learning module is used to learn the network parameters of the corresponding network layer in the AI ​​refined model according to the feature information fusion matrix and the guidance signal fusion matrix, and generate a refined product detection model.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the global learning method for product detection models described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the product detection model global learning method according to any one of claims 1 to 7 when executed.

Images