Biodegradable plastic identification method and system
Through deep learning technology, twin convolutional networks and U-Net networks are built to achieve efficient identification of biodegradable plastic labels, solve the separation and recycling problems in the existing technology, improve the identification accuracy and real-timeness, and support environmental protection and garbage classification Application in the field.
Patent Information
- Application Number
- CN202411491022.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-24
- Publication Date
- 2025-02-14
AI Technical Summary
The prior art is difficult to efficiently and accurately separate and recycle biodegradable plastic waste, resulting in environmental pollution and waste of resources.
Using deep learning technology, by building twin convolutional networks and U-Net networks, biodegradable plastic identification is recognized in real time, image preprocessing and feature extraction are realized, and recognition accuracy and efficiency are improved.
It significantly improves the identification accuracy and real-time nature of biodegradable plastic labels, enhances the generalization ability of the model, maintains stability and accuracy in a diverse environment, and supports applications in the fields of environmental protection and garbage classification.
Smart Images
Figure CN119445476A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of environmental science and resource utilization technology, and in particular relates to a biodegradable plastic identification method and system. Background Art
[0002] With the increasing demand and consumption of biodegradable plastic products, the environmental problems caused by biodegradable plastic waste have become a new challenge for my country's sustainable development. At present, most biodegradable plastic waste is mixed with other waste and enters waste incineration and landfill facilities, causing environmental damage no less than traditional plastic waste. Therefore, how to efficiently and accurately separate ordinary biodegradable plastics and recycle high-cost biodegradable materials is an urgent problem to be solved.
[0003] Traditional plastic identification methods mainly rely on manual judgment or simple image processing technology, which has problems such as low efficiency, low accuracy, and high cost. However, deep learning technology has achieved remarkable results in the field of image recognition and provided a new solution for the automatic identification of degradable garbage bag labels. Summary of the invention
[0004] In order to solve the above technical problems, the present invention proposes a biodegradable plastic identification method and system to solve the problems existing in the above prior art.
[0005] To achieve the above object, the present invention provides a method for identifying a biodegradable plastic mark, comprising:
[0006] Collect images with biodegradable plastic labels and images without labels based on the database;
[0007] Preprocessing the images with the biodegradable plastic logo and the images without the logo to obtain a preprocessed data set;
[0008] Constructing two identical convolutional neural networks, constructing a twin convolutional network based on the two identical convolutional neural networks, and training the twin convolutional network with the preprocessed data set to obtain a logo recognition model;
[0009] The plastic image is acquired in real time, pre-processed based on the U-Net network, and then input into the logo recognition model to generate a degradable plastic logo recognition result.
[0010] Preferably, the process of obtaining the preprocessed data set includes:
[0011] Cropping the image with the biodegradable plastic logo and the image without the logo to obtain a cropped image;
[0012] Rotating, scaling, shearing and color adjusting the cropped image to generate an enhanced data set;
[0013] Generate binary classification labels based on the attributes of the enhanced data set, and feed the binary classification labels back to the enhanced data set to generate a label data set;
[0014] Perform image processing on the label data set to obtain the preprocessed data set.
[0015] Preferably, the process of performing image processing on the label data set includes:
[0016] The label data set is converted into a grayscale image, and the grayscale image is subjected to denoising in sequence based on an adaptive median filtering algorithm to smooth the image and eliminate salt and pepper noise to obtain a denoised image;
[0017] Performing histogram equalization processing on the denoised image to generate an equalized image;
[0018] Performing geometric correction on the equalized image through perspective transformation to generate a corrected image;
[0019] Constructing a multi-scale Gabor filter group, extracting texture features of the wear area in the corrected image based on the multi-scale Gabor filter group, and obtaining a Gabor texture feature map;
[0020] The Gabor texture feature map is binarized, and the binarization threshold is adaptively determined using the Otsu threshold segmentation algorithm to generate the preprocessed data set.
[0021] Preferably, the process of constructing a twin convolutional network based on the two identical convolutional neural networks includes:
[0022] Build a spatially scaled attention mechanism and two identical convolutional neural networks;
[0023] The spatial scaling attention mechanism is connected to the two identical convolutional neural networks to generate two identical optimized convolutional neural networks.
[0024] Preferably, the expression of the spatial scaling attention mechanism is:
[0025]
[0026] Among them, Concat() is a concatenation operation. is the weight parameter, N i is the number of attention heads in the i-th stage, so the dimension of each head is LSR() performs a linear dimensionality reduction operation on the spatial dimension of the input sequence.
[0027] Preferably, the process of training the twin convolutional network through the preprocessed data set to obtain the identification model includes:
[0028] Obtaining a preliminary feature representation of the preprocessed data set based on a feature cascade method;
[0029] For the preliminary feature representation, a weight optimization algorithm is used to determine the weight coefficient of each feature to obtain an optimized feature representation;
[0030] Inputting the optimized feature representation into the twin convolutional network, and obtaining the deep features of the degradable marker through model training;
[0031] The goodness of fit and prediction accuracy of the deep features of the degradable marker are calculated based on a multivariate nonlinear regression model, and the trained twin convolutional network is optimized based on the calculation results to generate the marker recognition model.
[0032] Preferably, the loss function of the twin convolutional network is expressed as:
[0033]
[0034] Among them, Y is the label, D is the distance between two feature vectors, and m is the boundary value.
[0035] Preferably, the process of acquiring the plastic image in real time and preprocessing the plastic image based on the U-Net network includes:
[0036] Using a high-definition camera to collect images of plastic products at multiple angles and under multiple lighting conditions, to obtain collected images;
[0037] Filtering the collected image by using a Laplace operator to obtain a filtering result, and classifying the collected image based on the filtering result to generate a clear image and an unclear image;
[0038] The blurred area of the non-clear image is reconstructed based on the U-Net network to obtain a preprocessed image.
[0039] To achieve the above object, the present invention provides a biodegradable plastic identification system, comprising:
[0040] An image acquisition module, used for collecting images with biodegradable plastic labels and images without labels based on a database;
[0041] An image preprocessing module, used for preprocessing the images with biodegradable plastic labels and the images without labels to obtain a preprocessed data set;
[0042] A deep learning model module, used to construct two identical convolutional neural networks, construct a twin convolutional network based on the two identical convolutional neural networks, train the twin convolutional network through the preprocessed data set, and obtain a logo recognition model;
[0043] The result output module is used to obtain plastic images in real time, pre-process the plastic images based on the U-Net network, and then input them into the logo recognition model to generate degradable plastic logo recognition results.
[0044] Compared with the prior art, the present invention has the following advantages and technical effects:
[0045] The present invention collects images with biodegradable plastic labels and images without labels, performs data preprocessing, and enhances the generalization ability of the model. Data enhancement techniques such as rotation, scaling, and cropping are used to ensure that the model can maintain stability and accuracy in a variety of environments. The present invention constructs two identical convolutional neural networks and fuses them into a twin convolutional network. This structure can effectively compare the similarities between input images and help the model learn to recognize the subtle differences between images with degradable labels and images without labels. In the real-time processing stage, the present invention uses a U-Net network to preprocess the input plastic image to improve the image quality so that the label recognition model can better extract features. After processing, the present invention can quickly generate recognition results for degradable plastic labels. Overall, the present invention significantly improves the accuracy and real-time performance of label recognition through deep learning and image processing. The present invention has broad application potential in the fields of environmental protection, garbage classification, etc., which helps to reduce plastic pollution and promote the realization of sustainable development goals. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0047] Figure 1 This is a flow chart of a biodegradable plastic identification method according to an embodiment of the present invention;
[0048] Figure 2 Schematic diagram of a biodegradable plastic identification system according to an embodiment of the present invention. DETAILED DESCRIPTION
[0049] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0050] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0051] Embodiment 1
[0052] like Figure 1 As shown, this embodiment provides a method for identifying a biodegradable plastic mark, comprising:
[0053] Collect images with biodegradable plastic labels and images without labels based on the database;
[0054] Preprocessing the images with the biodegradable plastic logo and the images without the logo to obtain a preprocessed data set;
[0055] Constructing two identical convolutional neural networks, constructing a twin convolutional network based on the two identical convolutional neural networks, and training the twin convolutional network with the preprocessed data set to obtain a logo recognition model;
[0056] The plastic image is acquired in real time, pre-processed based on the U-Net network, and then input into the logo recognition model to generate a degradable plastic logo recognition result.
[0057] Specifically:
[0058] Image dataset: Collect images with and without biodegradable plastic logos. Make sure the dataset contains a variety of different logo styles and material types to enhance the generalization ability of the model.
[0059] The image data set includes degradable plastic bags, degradable tableware, degradable packaging materials, etc.;
[0060] Each image in the dataset is labeled to indicate whether it is a biodegradable plastic. A binary classification label can be used (e.g. 1 for biodegradable and 0 for non-biodegradable).
[0061] Data enhancement is performed to improve the robustness of the model, such as rotation, scaling, flipping, brightness adjustment, etc. Data enhancement can help the model learn more features.
[0062] Two identical convolutional neural networks (CNNs) are constructed, which share weights. The two networks receive two input images (e.g., the image to be recognized and the reference image).
[0063] The output of each sub-network passes through a series of convolutional layers, pooling layers, and activation layers to extract image features.
[0064] The trained model is applied to a new image dataset for real-time recognition of biodegradable plastic logos.
[0065] Compare the output feature vectors of the two sub-networks. Commonly used comparison methods include Euclidean distance, cosine similarity, etc. to calculate the similarity between the two feature vectors.
[0066] Further optimizing the scheme, the process of obtaining the preprocessed data set includes:
[0067] Cropping the image with the biodegradable plastic logo and the image without the logo to obtain a cropped image;
[0068] Rotating, scaling, shearing and color adjusting the cropped image to generate an enhanced data set;
[0069] Generate binary classification labels based on the attributes of the enhanced data set, and feed the binary classification labels back to the enhanced data set to generate a label data set;
[0070] Perform image processing on the label data set to obtain the preprocessed data set.
[0071] Image preprocessing is performed to improve image quality. A Gaussian filter can be used for denoising. Assuming that the noise of the image is Gaussian noise, a 3x3 Gaussian kernel is used for convolution operation to smooth the image. Next, histogram equalization is performed to enhance the contrast of the image, making the wear area more obvious. Then, a multi-scale Gabor filter group is constructed to extract texture features of different scales and directions. Assuming that a Gabor filter group of 5 scales and 8 directions is used, the image is convolved to obtain 40 response maps. Each response map represents the texture features of the image at a specific scale and direction. Next, the grayscale, texture and edge features of the wear area of the degradable marker are extracted. The grayscale feature can be obtained by calculating the average grayscale value of the wear area; the texture feature is statistically analyzed through the response map of the Gabor filter, such as calculating the mean and standard deviation of the response map; the edge feature can be extracted by the Canny edge detection algorithm, setting the low threshold to 50 and the high threshold to 150 to obtain the edge map. These features are combined into a feature vector. According to the extracted feature vector, the support vector machine algorithm is used to train the wear area recognition model. 1,000 labeled degradable logo images were used as training sets, and the RBF kernel function was used to set the penalty parameter C to 1 for model training. The trained model was used to determine the location and shape of the wear area of the degradable logo image to be tested. If the wear area was identified, the degree of wear of the degradable logo was determined based on its location and shape information.
[0072] Further optimizing the scheme, the process of image processing the labeled data set includes:
[0073] The label data set is converted into a grayscale image, and the grayscale image is subjected to denoising in sequence based on an adaptive median filtering algorithm to smooth the image and eliminate salt and pepper noise to obtain a denoised image;
[0074] Performing histogram equalization processing on the denoised image to generate an equalized image;
[0075] Performing geometric correction on the equalized image through perspective transformation to generate a corrected image;
[0076] Constructing a multi-scale Gabor filter group, extracting texture features of the wear area in the corrected image based on the multi-scale Gabor filter group, and obtaining a Gabor texture feature map;
[0077] The Gabor texture feature map is binarized, and the binarization threshold is adaptively determined using the Otsu threshold segmentation algorithm to generate the preprocessed data set.
[0078] Specifically:
[0079] Step 1: Obtain the degradable logo data and convert it into a grayscale image. Step 2: Use the adaptive median filtering algorithm to denoise the grayscale image, smooth the image, eliminate the salt and pepper noise, and obtain the denoised image. Step 3: Perform histogram equalization on the denoised image, stretch the pixel grayscale value distribution range, and enhance the overall image contrast. Step 4: Perform geometric correction on the equalized image through perspective transformation to eliminate the distortion introduced by lens distortion and restore the original shape of the degradable logo. Step 5: On the corrected image, design a multi-scale Gabor filter group based on the grayscale, texture, and edge features of the degradable logo wear area to extract the texture features of the wear area. Step 6: Binarize the Gabor texture feature map, use the Otsu threshold segmentation algorithm to adaptively determine the binarization threshold, and segment the wear area. Step 7: For the segmented wear area, extract its shape, area, grayscale statistical histogram and other feature vectors for subsequent judgment and classification of the degradable logo.
[0080] Further optimizing the scheme, the process of constructing a twin convolutional network based on the two identical convolutional neural networks includes:
[0081] Build a spatially scaled attention mechanism and two identical convolutional neural networks;
[0082] The spatial scaling attention mechanism is connected to the two identical convolutional neural networks to generate two identical optimized convolutional neural networks.
[0083] Further optimizing the scheme, the expression of the spatial scaling attention mechanism is:
[0084]
[0085] Among them, Concat() is a concatenation operation. is the weight parameter, N i is the number of attention heads in the i-th stage, so the dimension of each head is LSR() performs a linear dimensionality reduction operation on the spatial dimension of the input sequence.
[0086] Further optimizing the scheme, the process of training the twin convolutional network through the preprocessed data set to obtain the identification model includes:
[0087] Obtaining a preliminary feature representation of the preprocessed data set based on a feature cascade method;
[0088] For the preliminary feature representation, a weight optimization algorithm is used to determine the weight coefficient of each feature to obtain an optimized feature representation;
[0089] Inputting the optimized feature representation into the twin convolutional network, and obtaining the deep features of the degradable marker through model training;
[0090] The goodness of fit and prediction accuracy of the deep features of the degradable marker are calculated based on a multivariate nonlinear regression model, and the trained twin convolutional network is optimized based on the calculation results to generate the marker recognition model.
[0091] On the basis of the twin network, a multivariate nonlinear regression model can be introduced to further process the feature vector. This regression model can be one or more fully connected layers, which can learn the nonlinear relationship between the feature vector and the label. In this way, the network can not only learn the similarities between the inputs, but also predict the specific relationship or difference between them.
[0092] The parameter architecture of the twin convolutional network is:
[0093] Input Layer:
[0094] The input size is 224x224x3, suitable for color images.
[0095] Shared convolutional layers:
[0096] Convolutional layer 1: 32 filters, each filter size is 3x3, and the activation function uses ReLU.
[0097] Pooling layer 1: 2x2 max pooling.
[0098] Convolutional layer 2: 64 filters, filter size is 3x3, and the activation function uses ReLU.
[0099] Pooling layer 2: 2x2 max pooling.
[0100] Convolutional layer 3: 128 filters, filter size is 3x3, and activation function uses ReLU.
[0101] Pooling layer 3: 2x2 max pooling.
[0102] Flattening layer: Flattens the feature map into a one-dimensional vector.
[0103] Fully connected layers: Each branch uses a fully connected layer with ReLU activation and an output dimension of 256.
[0104] Similarity calculation: Calculate the L1 distance between the outputs of two fully connected layers, and then convert this distance into a similarity score through a fully connected layer with sigmoid activation.
[0105] Training parameters
[0106] Loss function: For binary classification problems, use contrastive loss or cross entropy loss.
[0107] Optimizer: Adam, the learning rate is initially set to 0.001.
[0108] Batch Size: Select 32 or 64.
[0109] Iteration number: gradually adjusted according to performance, usually ranging from 10-50 times.
[0110] Further optimizing the scheme, the loss function of the twin convolutional network is expressed as:
[0111]
[0112] Among them, Y is the label, D is the distance between two feature vectors, and m is the boundary value.
[0113] The scheme is further optimized to obtain plastic images in real time. The process of preprocessing the plastic images based on the U-Net network includes:
[0114] Using a high-definition camera to collect images of plastic products at multiple angles and under multiple lighting conditions, to obtain collected images;
[0115] Filtering the collected image by using a Laplace operator to obtain a filtering result, and classifying the collected image based on the filtering result to generate a clear image and an unclear image;
[0116] Based on the U-Net network structure, this embodiment performs in-depth analysis and processing on the blurred area in the input non-clear image, and effectively extracts the feature information of the image through the encoder-decoder architecture of the network, and accurately reconstructs the blurred area. In the encoding stage, the network gradually extracts the low-level features of the image, and gradually restores the high-level features of the image in the decoding stage, while enhancing the blurred area and reconstructing the details.
[0117] Embodiment 2
[0118] like Figure 2 As shown, this embodiment provides a biodegradable plastic identification system, including:
[0119] An image acquisition module, used for collecting images with biodegradable plastic labels and images without labels based on a database;
[0120] An image preprocessing module, used for preprocessing the images with biodegradable plastic labels and the images without labels to obtain a preprocessed data set;
[0121] A deep learning model module, used to construct two identical convolutional neural networks, construct a twin convolutional network based on the two identical convolutional neural networks, train the twin convolutional network through the preprocessed data set, and obtain a logo recognition model;
[0122] The result output module is used to obtain plastic images in real time, pre-process the plastic images based on the U-Net network, and then input them into the logo recognition model to generate degradable plastic logo recognition results.
[0123] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A method for identifying a biodegradable plastic mark, characterized in that: The following steps are involved: Collect images with biodegradable plastic labels and images without labels based on the database; Preprocessing the images with the biodegradable plastic logo and the images without the logo to obtain a preprocessed data set; Constructing two identical convolutional neural networks, constructing a twin convolutional network based on the two identical convolutional neural networks, and training the twin convolutional network with the preprocessed data set to obtain a logo recognition model; The plastic image is acquired in real time, pre-processed based on the U-Net network, and then input into the logo recognition model to generate a degradable plastic logo recognition result.
2. The biodegradable plastic identification method according to claim 1, characterized in that: The process of obtaining the preprocessed data set includes: Cropping the image with the biodegradable plastic logo and the image without the logo to obtain a cropped image; Rotating, scaling, shearing and color adjusting the cropped image to generate an enhanced data set; Generate binary classification labels based on the attributes of the enhanced data set, and feed the binary classification labels back to the enhanced data set to generate a label data set; Perform image processing on the label data set to obtain the preprocessed data set.
3. The biodegradable plastic identification method according to claim 2, characterized in that: The process of performing image processing on the label data set includes: The label data set is converted into a grayscale image, and the grayscale image is subjected to denoising in sequence based on an adaptive median filtering algorithm to smooth the image and eliminate salt and pepper noise to obtain a denoised image; Performing histogram equalization processing on the denoised image to generate an equalized image; Performing geometric correction on the equalized image through perspective transformation to generate a corrected image; Constructing a multi-scale Gabor filter group, extracting texture features of the wear area in the corrected image based on the multi-scale Gabor filter group, and obtaining a Gabor texture feature map; The Gabor texture feature map is binarized, and the binarization threshold is adaptively determined using the Otsu threshold segmentation algorithm to generate the preprocessed data set.
4. The biodegradable plastic identification method according to claim 1, characterized in that: The process of constructing a twin convolutional network based on the two identical convolutional neural networks includes: Build a spatially scaled attention mechanism and two identical convolutional neural networks; The spatial scaling attention mechanism is connected to the two identical convolutional neural networks to generate two identical optimized convolutional neural networks.
5. The biodegradable plastic identification method according to claim 4, characterized in that: The expression of the spatial scaling attention mechanism is: Among them, Concat() is a concatenation operation. is the weight parameter, N i is the number of attention heads in the i-th stage, so the dimension of each head is LSR() performs a linear dimensionality reduction operation on the spatial dimension of the input sequence.
6. The biodegradable plastic identification method according to claim 1, characterized in that: The process of training the twin convolutional network through the preprocessed data set to obtain the logo recognition model includes: Obtaining a preliminary feature representation of the preprocessed data set based on a feature cascade method; For the preliminary feature representation, a weight optimization algorithm is used to determine the weight coefficient of each feature to obtain an optimized feature representation; Inputting the optimized feature representation into the twin convolutional network, and obtaining the deep features of the degradable marker through model training; The goodness of fit and prediction accuracy of the deep features of the degradable marker are calculated based on a multivariate nonlinear regression model, and the trained twin convolutional network is optimized based on the calculation results to generate the marker recognition model.
7. The biodegradable plastic identification method according to claim 1, characterized in that: The loss function of the twin convolutional network is expressed as: Among them, Y is the label, D is the distance between two feature vectors, and m is the boundary value.
8. The biodegradable plastic identification method according to claim 1, characterized in that: The process of acquiring a plastic image in real time and preprocessing the plastic image based on a U-Net network includes: Using a high-definition camera to collect images of plastic products at multiple angles and under multiple lighting conditions, to obtain collected images; Filtering the collected image by using a Laplace operator to obtain a filtering result, and classifying the collected image based on the filtering result to generate a clear image and an unclear image; The blurred area of the non-clear image is reconstructed based on the U-Net network to obtain a preprocessed image.
9. The system of the biodegradable plastic identification method according to any one of claims 1 to 8, characterized in that: include: An image acquisition module, used for collecting images with biodegradable plastic labels and images without labels based on a database; An image preprocessing module, used for preprocessing the images with biodegradable plastic labels and the images without labels to obtain a preprocessed data set; A deep learning model module, used to construct two identical convolutional neural networks, construct a twin convolutional network based on the two identical convolutional neural networks, train the twin convolutional network through the preprocessed data set, and obtain a logo recognition model; The result output module is used to obtain plastic images in real time, pre-process the plastic images based on the U-Net network, and then input them into the logo recognition model to generate degradable plastic logo recognition results.