Waste household appliance recycling and classifying method and system based on machine vision

Abstract

The invention discloses a waste household appliance recycling and classifying method and system based on machine vision, and the method comprises the steps: S1, enhancing a waste household appliance part image through employing a multi-scale Retinex method, and obtaining an enhanced image; s2, extracting local texture features of the enhanced image by using a local binary pattern method to obtain a texture feature map; s3, constructing a waste household appliance part classification model based on VGG-16, receiving the enhanced image and the textural feature map, and outputting large-class and small-class probabilities of the waste household appliance parts; s4, constructing a multi-task loss function to train the waste household appliance part classification model to obtain a trained model; and S5, by using the trained waste household appliance part classification model, inputting the unlabeled enhanced image and the textural feature map to obtain subclass probability distribution, and determining a final classification result. The waste household electrical appliance recycling and classifying solution provided by the invention is high in automation degree, accurate in classification and high in processing speed.

Description

Technical Field

[0001] The present invention relates to the technical field of waste household appliances recycling, and in particular to a waste household appliances recycling classification method and system based on machine vision. Background Art

[0002] With the advancement of science and technology and the development of society, the recycling of used household appliances has become an important environmental issue. Traditional methods for recycling and sorting used household appliances mostly rely on manual identification, which is not only inefficient but also easily affected by personal experience and skills, resulting in unstable classification accuracy. In addition, due to the wide variety and different shapes of used household appliance parts, traditional manual identification methods are difficult to cope with complex parts identification tasks.

[0003] In recent years, with the development of machine vision technology, more and more studies have begun to try to use computer vision technology to solve the problem of automatic classification of waste home appliance parts. However, existing methods face many challenges in processing images of waste home appliance parts. The images of waste home appliance parts are often affected by factors such as lighting conditions and shooting angles, resulting in low image quality and difficulty in extracting effective feature information from them. Existing feature extraction methods may not be able to effectively capture the key features of waste home appliance parts, especially in terms of details and textures, which makes subsequent classification tasks difficult. Since there may be great similarities between the major and minor categories of waste home appliance parts, traditional classification models are prone to misjudgment and have low classification accuracy. Summary of the invention

[0004] In view of this, the present invention provides a waste household appliance recycling and classification method based on machine vision, aiming to provide a waste household appliance recycling and classification solution with high automation, accurate classification and fast processing speed.

[0005] To achieve the above object, the present invention provides a method for recycling and classifying waste household appliances based on machine vision, comprising the following steps: S1: Use the multi-scale Retinex method to enhance the image of waste home appliance parts and obtain the enhanced image; S2: extract local texture features of the enhanced image using the local binary pattern method to obtain a texture feature map; S3: Build a classification model for used home appliance parts based on VGG-16 to receive the enhanced image and texture feature map, and output the major and minor category probabilities of used home appliance parts; S4: Construct a multi-task loss function to train the classification model of used household appliance parts to obtain a trained model; S5: Using the trained model for the classification of used household appliance parts, input the unlabeled enhanced image and texture feature map to obtain the sub-category probability distribution and determine the final classification result.

[0006] As a further improvement method of the present invention: Optionally, in step S1, the multi-scale Retinex method is used to enhance the image of the waste household appliance parts to obtain the enhanced image, including: The contrast and details of the image of the waste household appliance parts are enhanced using a multi-scale Retinex method, wherein the multi-scale Retinex method is specifically as follows: S11: Choose different scale parameters , , as the standard deviation of the Gaussian kernel; for each scale Use scale Gaussian filter Image of used home appliance parts Filter and get the scale Lighting estimation , specifically: in, is the radius of the Gaussian filter; and Images of used home appliance parts Pixel coordinates in horizontal and vertical directions; and Respectively represent the Gaussian filter relative to the center point The horizontal and vertical offsets of the pixel position, , ; Images of used home appliance parts At pixel coordinates The pixel value at ; For scale Gaussian filter The horizontal and vertical offsets are and The filter weight when ; For scale Lighting estimation At pixel coordinates The pixel value at ; is the ratio of pi; is a natural constant; S12: Calculation scale Reflectivity And normalize it, specifically: in, For scale Reflectivity At pixel coordinates The reflectivity value at ; Images of used home appliance parts At pixel coordinates The pixel value at ; ; To scale Reflectivity Perform normalization to obtain the normalized scale Reflectivity ; S13: Take the average of the normalized reflectance at different scales to obtain the comprehensive reflectance, and restore the comprehensive reflectance to the image of the waste home appliance parts At the brightness level, the enhanced image is obtained, and the comprehensive reflectivity Specifically: in, is the comprehensive reflectivity At pixel coordinates The value of the reflectivity at Multiply by the average brightness to get the enhanced image .

[0007] Optionally, in step S2, extracting local texture features of the enhanced image using a local binary pattern method to obtain a texture feature map includes: S21: For the enhanced image , select a pixel value of a central pixel And define the pixel values ​​of the 8 neighboring pixel positions of the central pixel , the neighborhood pixel position is located at the center pixel Within the range, they are numbered in counter-clockwise order; S22: Calculate local texture feature values ​​and generate texture feature maps: S221: Calculate the pixel value of the central pixel The pixel value of its neighboring pixel location , , the binarization result between , specifically: S222: Binarization result corresponding to the pixel value of each neighborhood pixel position Convert to an 8-bit binary number and convert it to decimal form to get the center pixel position The local texture eigenvalue at ; S223: For each pixel position , calculated Directly use it as the local texture feature value of the pixel position, and obtain the texture feature map after traversing all pixel positions .

[0008] Optionally, in step S3, a classification model for used household appliance parts is constructed based on VGG-16 to receive the enhanced image and texture feature map, and output the major and minor category probabilities of used household appliance parts, including: S31: Merge the enhanced image and texture feature map in the channel dimension of the image, specifically: in, is the merged input data; S32: Use the first 13 layers of the VGG-16 network, including convolutional layers and pooling layers, to extract image features, specifically: in, The first 13 layers of the VGG-16 network; is the extracted image features; S33: Input the extracted image features into two independent fully connected layers to obtain the classification results of the categories to which the waste home appliance parts belong And the classification results of small categories ; S34: Use the softmax function to convert the output of the fully connected layer into a probability distribution, specifically: in, To predict the used home appliance parts belonging to the first The probability of a large class; and They are No. Peace Dimensional results, , is the number of major categories, for dimensional vector; To predict the used home appliance parts belonging to the first The probability of a small class; and They are No. Peace Dimensional results, , is the number of sub-categories, for dimensional vector.

[0009] Optionally, in step S4, constructing a multi-task loss function to train the waste home appliance parts classification model to obtain a trained model includes: S41: Treat each merged input data input into the waste home appliance parts classification model as a training sample and define the real major category labels of the training samples and a real subclass label ,in, , ; S42: Use cross entropy loss to calculate and weight the large category classification loss and small category classification loss to obtain weighted loss , specifically: in, is the true label after one-hot encoding Vector elements, if the true category label of the training sample is ,but ,otherwise ; is the true label after one-hot encoding Vector elements, if the true small class label of the training sample is ,but ,otherwise ; S43: Calculate the average of the weighted losses of all training samples to obtain the total loss , specifically: in, is the total number of training samples; For the The weighted loss of samples; Stochastic gradient descent is used to iteratively reduce the total loss. After the number of iterations reaches the set number, the trained model for the classification of used household appliance parts is obtained.

[0010] Optionally, in step S5, the trained waste home appliance parts classification model is used to input the unlabeled enhanced image and texture feature map to obtain the sub-class probability distribution and determine the final classification result, including: The unlabeled image of the waste appliance parts is processed by the steps S1 and S2 to obtain the enhanced image and texture feature map, which is then input into the trained waste appliance parts classification model to obtain the probability of the sub-category. , select the small class with the largest probability as the classification result, specifically: in, The subcategory label of the unlabeled used appliance parts image predicted by the trained used appliance parts classification model.

[0011] The present invention also discloses a waste household appliance recycling and classification system based on machine vision, comprising: Enhancement module: Use multi-scale Retinex method to enhance the images of waste home appliance parts; Texture extraction module: uses the local binary pattern method to extract local texture features of the enhanced image; Network construction module: Build a classification model for used home appliance parts based on VGG-16 to receive enhanced images and texture feature maps, and output the probabilities of major and minor categories of used home appliance parts; Network training module: Construct a multi-task loss function to train the classification model of used household appliance parts; Network application module: Utilize the trained model for the classification of used household appliance parts, input the unlabeled enhanced image and texture feature map to obtain the sub-category probability distribution, and determine the final classification result.

[0012] Beneficial effects: The present invention enhances the contrast and details of the images of waste household appliance parts through a multi-scale Retinex method, effectively solving the problem that traditional image processing methods are difficult to extract effective features under complex lighting conditions. The multi-scale Retinex method can enhance the contrast of the image at different levels of detail through illumination estimation and reflectivity calculation at multiple scales, thereby making the key features in the image more prominent. This image enhancement technology significantly improves the accuracy of subsequent feature extraction and classification links. In addition, the local binary pattern, as a powerful texture descriptor, can capture local texture features in the image, further enriching the feature expression capability. Through the combination of these two key technologies, the present invention can achieve higher classification accuracy in complex waste household appliance parts classification tasks, reduce the possibility of misclassification, and improve the reliability and robustness of the overall system.

[0013] Traditional classification of used household appliance parts relies on manual identification, which is not only time-consuming and labor-intensive, but also easily affected by individual differences, resulting in unstable classification results. The present invention utilizes advanced machine vision technology and deep learning models to achieve efficient and automated classification of used household appliance parts. The deep learning model based on VGG-16 can automatically learn multi-level feature representations from images. By training the model to distinguish between different major and minor categories of used household appliance parts, the present invention greatly reduces the need for manual intervention and makes the classification process more efficient and quick. In addition, the design of the multi-task loss function enables the model to simultaneously optimize the classification performance of major and minor categories during the training process, further improving the efficiency of classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The present invention is a flowchart of a method for recycling and classifying waste household appliances based on machine vision according to an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0016] Embodiment 1: A method for recycling and classifying waste household appliances based on machine vision, such as Figure 1 As shown, the following steps are included: S1: Use the multi-scale Retinex method to enhance the image of waste home appliance parts and obtain the enhanced image: The contrast and details of the image of the waste household appliance parts are enhanced using a multi-scale Retinex method, wherein the multi-scale Retinex method is specifically as follows: S11: Choose different scale parameters , , as the standard deviation of the Gaussian kernel; for each scale Use scale Gaussian filter Image of used home appliance parts Filter and get the scale Lighting estimation , specifically: in, is the radius of the Gaussian filter; and Images of used home appliance parts Pixel coordinates in horizontal and vertical directions; and Respectively represent the Gaussian filter relative to the center point The horizontal and vertical offsets of the pixel position, , ; Images of used home appliance parts At pixel coordinates The pixel value at ; For scale Gaussian filter The horizontal and vertical offsets are and The filter weight when ; For scale Lighting estimation At pixel coordinates The pixel value at ; is the ratio of pi; is a natural constant; S12: Calculation scale Reflectivity And normalize it, specifically: in, For scale Reflectivity At pixel coordinates The reflectivity value at ; Images of used home appliance parts At pixel coordinates The pixel value at ; ; To scale Reflectivity Normalization is performed, specifically: in, is the normalized scale Reflectivity; is the normalized scale Reflectivity At pixel coordinates The reflectivity value at ; For scale Reflectivity The minimum value among all pixel coordinates; For scale Reflectivity on The maximum value among all pixel coordinates; S13: Take the average of the normalized reflectance at different scales to obtain the comprehensive reflectance, and restore the comprehensive reflectance to the image of the waste home appliance parts At the brightness level, the enhanced image is obtained, and the comprehensive reflectivity Specifically: in, is the comprehensive reflectivity At pixel coordinates The value of the reflectivity at Multiply by the average brightness to get the enhanced image , specifically: in, For the enhanced image At pixel coordinates The pixel value at ; Images of used home appliance parts The mean pixel value of .

[0017] The multi-scale Retinex method can enhance the contrast and details of images at different scales, which is especially important for images of used household appliance parts. Used household appliance parts usually have complex structures and textures, and may be photographed under different lighting conditions. Traditional image enhancement methods often only work at a single scale and may ignore some subtle but important features. The multi-scale Retinex method can enhance images at different levels of detail by performing illumination estimation and reflectivity calculation at multiple scales, making the key features in the image more obvious. This method can effectively improve the overall contrast of the image and make the details clearer, thereby providing high-quality input data for subsequent feature extraction and classification.

[0018] S2: Use the local binary pattern method to extract the local texture features of the enhanced image and obtain the texture feature map: S21: For the enhanced image , select a pixel value of a central pixel And define the pixel values ​​of the 8 neighboring pixel positions of the central pixel , the neighborhood pixel position is located at the center pixel Within the range, they are numbered in counter-clockwise order; S22: Calculate local texture feature values ​​and generate texture feature maps: S221: Calculate the pixel value of the central pixel The pixel value of its neighboring pixel location , , the binarization result between , specifically: S222: Binarization result corresponding to the pixel value of each neighborhood pixel position Convert to an 8-bit binary number and convert it to decimal form to get the center pixel position The local texture eigenvalue at , specifically: S223: For each pixel position , calculated Directly use it as the local texture feature value of the pixel position, and obtain the texture feature map after traversing all pixel positions .

[0019] The local binary pattern method is very simple and computationally efficient. It only requires comparing the intensity values ​​of the central pixel with the surrounding pixels and encoding the comparison result into a numerical value. The local binary pattern feature has good rotation invariance. When the image is rotated, the change in the local texture feature value is relatively small, because the local binary pattern feature focuses on the relationship between the central pixel and the surrounding pixels, rather than the absolute pixel value. This property enables the local binary pattern feature to maintain good recognition performance when processing rotationally changed images.

[0020] S3: Build a classification model for used home appliance parts based on VGG-16 to receive the enhanced image and texture feature map, and output the major and minor category probabilities of used home appliance parts: S31: Merge the enhanced image and texture feature map in the channel dimension of the image, specifically: in, is the merged input data; S32: Use the first 13 layers of the VGG-16 network, including convolutional layers and pooling layers, to extract image features, specifically: in, The first 13 layers of the VGG-16 network; is the extracted image features; S33: The extracted image features are input into two independent fully connected layers for classification of the major and minor categories of waste home appliance parts, specifically: in, and Represents the results of the two independent fully connected layers outputting the used home appliance parts belonging to different categories and subcategories; and They are the weights and biases of the fully connected layer for large-category classification; and are the weights and biases of the fully connected layer for small category classification; The extracted image features The eigenvector obtained when the mapping is flattened into a one-dimensional vector; S34: Use the softmax function to convert the output of the fully connected layer into a probability distribution, specifically: in, To predict the used home appliance parts belonging to the first The probability of a large class; and They are No. Peace Dimensional results, , is the number of major categories, for dimensional vector; To predict the used home appliance parts belonging to the first The probability of a small class; and They are No. Peace Dimensional results, , is the number of sub-categories, for dimensional vector.

[0021] By using the first 13 layers of the VGG-16 network to extract image features, we can take advantage of the powerful representation learning capabilities of deep learning. VGG-16 is a classic deep learning architecture that performs well in image classification tasks. Its convolutional and pooling layers can automatically learn multi-level abstract features in images, which helps improve classification accuracy.

[0022] By inputting image features into two independent fully connected layers to classify large and small categories, multi-task learning is achieved. The model in the present invention can simultaneously learn how to distinguish different large categories and subdivided small categories under the same category. This method can not only improve the generalization ability of the model, but also help the model better understand the hierarchical structure between waste home appliance parts.

[0023] S4: Construct a multi-task loss function to train the classification model of used household appliance parts and obtain the trained model: S41: Treat each merged input data input into the waste home appliance parts classification model as a training sample and define the real major category labels of the training samples and a real subclass label ,in, , ; S42: Use cross entropy loss to calculate and weight the large category classification loss and small category classification loss to obtain weighted loss , specifically: in, is the true label after one-hot encoding Vector elements, if the true category label of the training sample is ,but ,otherwise ; is the true label after one-hot encoding Vector elements, if the true small class label of the training sample is ,but ,otherwise ; S43: Calculate the average of the weighted losses of all training samples to obtain the total loss , specifically: in, is the total number of training samples; For the The weighted loss of samples; Stochastic gradient descent is used to iteratively reduce the total loss. After the number of iterations reaches the set number, the trained model for the classification of used household appliance parts is obtained.

[0024] By using a multi-task loss function, this step can simultaneously optimize the classification performance of large and small categories. This approach allows the model to learn large-category classification while also taking into account the classification requirements of small categories. This synergistic effect helps improve the overall performance of the model because there may be shared information between the two tasks, and mutual promotion can improve the accuracy of each task.

[0025] S5: Using the trained waste appliance parts classification model, input the unlabeled enhanced image and texture feature map to obtain the sub-class probability distribution and determine the final classification result: The unlabeled image of the waste appliance parts is processed by the steps S1 and S2 to obtain the enhanced image and texture feature map, which is then input into the trained waste appliance parts classification model to obtain the probability of the sub-category. , select the small class with the largest probability as the classification result, specifically: in, The subcategory label of the unlabeled used appliance parts image predicted by the trained used appliance parts classification model.

[0026] Embodiment 2: The present invention also discloses a waste household appliance recycling and classification system based on machine vision, which includes the following five modules: Enhancement module: Use multi-scale Retinex method to enhance the images of waste home appliance parts; Texture extraction module: uses the local binary pattern method to extract local texture features of the enhanced image; Network construction module: Build a classification model for used home appliance parts based on VGG-16 to receive enhanced images and texture feature maps, and output the probabilities of major and minor categories of used home appliance parts; Network training module: Construct a multi-task loss function to train the classification model of used household appliance parts; Network application module: Utilize the trained model for the classification of used household appliance parts, input the unlabeled enhanced image and texture feature map to obtain the sub-category probability distribution, and determine the final classification result.

[0027] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0028] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0029] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for recycling and classifying waste household appliances based on machine vision, characterized in that: The following steps are involved: S1: Use the multi-scale Retinex method to enhance the image of waste home appliance parts and obtain the enhanced image; S2: extract local texture features of the enhanced image using the local binary pattern method to obtain a texture feature map; S3: Build a classification model for used home appliance parts based on VGG-16 to receive the enhanced image and texture feature map, and output the major and minor category probabilities of used home appliance parts; S4: Construct a multi-task loss function to train the classification model of used household appliance parts to obtain a trained model; S5: Using the trained waste appliance parts classification model, input the unlabeled enhanced image and texture feature map to obtain the sub-class probability distribution and determine the final classification result: After the unlabeled image of the waste appliance parts is processed by the steps S1 and S2 to obtain the enhanced image and texture feature map, it is input into the trained waste appliance parts classification model to obtain the probability of the sub-category. , select the small class with the largest probability as the classification result, specifically: in, The subcategory label of the unlabeled used appliance parts image predicted by the trained used appliance parts classification model.

2. The method for recycling and classifying waste household appliances based on machine vision according to claim 1 is characterized in that: The step S1 includes the following steps: Use multi-scale Retinex method to enhance the contrast and details of images of scrap home appliance parts; The multi-scale Retinex method is specifically: S11: Choose different scale parameters , , as the standard deviation of the Gaussian kernel; for each scale Use scale Gaussian filter Image of used home appliance parts Filter and get the scale Lighting estimation , specifically: in, is the radius of the Gaussian filter; and Images of used home appliance parts Pixel coordinates in horizontal and vertical directions; and Respectively represent the Gaussian filter relative to the center point The horizontal and vertical offsets of the pixel position, , ; Images of used home appliance parts At pixel coordinates The pixel value at ; For scale Gaussian filter The horizontal and vertical offsets are and The filter weight when ; For scale Lighting estimation At pixel coordinates The pixel value at ; is pi; is a natural constant; S12: Calculation scale Reflectivity And normalize it, specifically: in, For scale Reflectivity At pixel coordinates The reflectivity value at ; Images of used home appliance parts At pixel coordinates The pixel value at ; ; To scale Reflectivity Perform normalization to obtain the normalized scale Reflectivity ; S13: Take the average of the normalized reflectance at different scales to obtain the comprehensive reflectance, and restore the comprehensive reflectance to the image of the waste home appliance parts At the brightness level, the enhanced image is obtained, and the comprehensive reflectivity Specifically: in, is the comprehensive reflectivity At pixel coordinates The value of the reflectivity at Multiply by the average brightness to get the enhanced image .

3. The method for recycling and classifying waste household appliances based on machine vision according to claim 2 is characterized in that: The step S2 includes the following steps: S21: For the enhanced image , select a pixel value of a central pixel And define the pixel values ​​of the 8 neighboring pixel positions of the central pixel , the neighborhood pixel position is located at the center pixel Within the range, they are numbered in counter-clockwise order; S22: Calculate local texture feature values ​​and generate a texture feature map.

4. The method for recycling and classifying waste household appliances based on machine vision according to claim 3 is characterized in that: The step S22 includes the following steps: S221: Calculate the pixel value of the central pixel The pixel value of its neighboring pixel location , , the binarization result between , specifically: S222: Binarization result corresponding to the pixel value of each neighborhood pixel position Convert to an 8-bit binary number and convert it to decimal form to get the center pixel position The local texture eigenvalue at ; S223: For each pixel position , calculated Directly use it as the local texture feature value of the pixel position, and obtain the texture feature map after traversing all pixel positions .

5. The method for recycling and classifying waste household appliances based on machine vision according to claim 3 is characterized in that: The step S3 includes the following steps: S31: Merge the enhanced image and texture feature map in the channel dimension of the image, specifically: in, is the merged input data; S32: Use the first 13 layers of the VGG-16 network, including convolutional layers and pooling layers, to extract image features, specifically: in, The first 13 layers of the VGG-16 network; is the extracted image features; S33: Input the extracted image features into two independent fully connected layers to obtain the classification results of the categories to which the waste home appliance parts belong And the classification results of small categories ; S34: Use the softmax function to convert the output of the fully connected layer into a probability distribution, specifically: in, To predict the used home appliance parts belonging to the first The probability of a large class; and They are No. Peace Dimensional results, , is the number of major categories, for dimensional vector; To predict the used home appliance parts belonging to the first The probability of a small class; and They are No. Peace Dimensional results, , is the number of sub-categories, for dimensional vector.

6. The method for recycling and classifying waste household appliances based on machine vision according to claim 5 is characterized in that: The step S4 includes the following steps: S41: Treat each merged input data input into the waste home appliance parts classification model as a training sample and define the real major category labels of the training samples and a real subclass label ,in, , ; S42: Use cross entropy loss to calculate and weight the large category classification loss and small category classification loss to obtain weighted loss , specifically: in, is the true label after one-hot encoding Vector elements, if the true category label of the training sample is ,but ,otherwise ; is the true label after one-hot encoding Vector elements, if the true small class label of the training sample is ,but ,otherwise ; S43: Calculate the average of the weighted losses of all training samples to obtain the total loss , specifically: in, is the total number of training samples; For the The weighted loss of samples; Stochastic gradient descent is used to iteratively reduce the total loss. After the number of iterations reaches the set number, the trained model for the classification of used household appliance parts is obtained.

7. A waste household appliance recycling and classification system based on machine vision, characterized in that: include: Enhancement module: Use multi-scale Retinex method to enhance the images of waste home appliance parts; Texture extraction module: uses the local binary pattern method to extract local texture features of the enhanced image; Network construction module: Build a classification model for used home appliance parts based on VGG-16 to receive enhanced images and texture feature maps, and output the probabilities of major and minor categories of used home appliance parts; Network training module: Construct a multi-task loss function to train the classification model of used household appliance parts; Network application module: using the trained model for the classification of used household appliance parts, inputting unlabeled enhanced images and texture feature maps to obtain sub-class probability distribution and determine the final classification results; To realize a method for recycling and classifying used household appliances based on machine vision as described in any one of claims 1 to 6.

Images