Plastic waste detection method, device and equipment and storage medium

Abstract

The invention relates to the field of plastic waste detection, in particular to a plastic waste detection method, device and equipment and a storage medium. The method comprises the following steps: acquiring a multi-angle image of a to-be-detected area; performing multi-scale detail reconstruction on the multi-angle image of the region to be detected, and performing region background feature interference suppression so as to construct a background suppression optimization image; performing in-region target visual detection on the background suppression optimization image, and performing object material component mining so as to generate a material component feature of each object; deep semantic segmentation is carried out based on the material component features of each object, image impurity removal processing is carried out, and a non-plastic impurity removal optimized image is constructed; and performing plastic waste visual inspection on the non-plastic impurity removal optimized image, and performing bounding box instance division processing so as to extract a bounding box image of each plastic waste. According to the invention, efficient and accurate plastic waste detection and identification are realized.

Description

Technical Field

[0001] The present invention relates to the field of plastic waste detection, and in particular to a plastic waste detection method, device, equipment and storage medium. Background Art

[0002] With the increasing awareness of environmental protection and the implementation of sustainable development strategies, the recycling and treatment of plastic waste has become an environmental issue that needs to be solved urgently around the world. Plastic materials are widely used in various industries due to their lightness, durability and low cost. However, the problem of handling these materials after use has become increasingly prominent. The pollution of the environment by plastic waste, especially marine pollution, soil pollution and damage to the ecosystem, has attracted widespread attention from all walks of life. In the current recycling process, how to efficiently and accurately identify and classify different types of plastic waste has become the key to improving recycling efficiency and reducing environmental pollution.

[0003] Traditional methods for detecting plastic waste mainly rely on technologies such as manual classification and traditional sensors, but these methods often face problems such as low efficiency, large classification errors, and high labor costs. Manual classification is not only time-consuming, but also easily affected by human factors, resulting in inaccurate identification and sorting; and traditional sensors have limited ability to distinguish plastic types in complex environments and are easily interfered by other substances. Therefore, how to adopt a more intelligent, efficient and accurate method to solve the problem of detecting and classifying plastic waste has become a technical problem that the industry urgently needs to break through. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a plastic waste detection method, device, equipment and storage medium to solve at least one of the above technical problems.

[0005] To achieve the above object, the present invention provides a method for detecting plastic waste, comprising the following steps: Step S1: acquiring multi-angle images of the area to be detected; reconstructing multi-scale details of the multi-angle images of the area to be detected, and suppressing interference of regional background features, thereby constructing a background suppression optimized image; Step S2: Performing visual detection of targets within the region of the background suppression optimized image and mining the material composition of the object, thereby generating material composition features of each object; Step S3: Perform deep semantic segmentation based on the material composition characteristics of each object, and perform image impurity removal processing to construct a non-plastic impurity removal optimized image; Step S4: performing visual inspection of plastic waste on the non-plastic impurity rejection optimization image, and performing bounding box instance segmentation processing, thereby extracting a bounding box image of each plastic waste; Step S5: Perform dynamic component learning and iterative detection optimization on the bounding box image of each plastic waste to build an intelligent waste detection optimization model.

[0006] The present invention also provides a detection device for plastic waste, comprising: The image optimization module is used to obtain multi-angle images of the area to be detected; reconstruct multi-scale details of the multi-angle images of the area to be detected, and suppress the interference of regional background features, so as to construct a background suppression optimized image; The material composition recognition module is used to perform visual detection of targets within the background suppression optimized image and to mine the material composition of the object, thereby generating the material composition features of each object; The impurity removal module is used to perform deep semantic segmentation based on the material composition characteristics of each object, and to remove image impurities to construct an optimized image for non-plastic impurity removal; A bounding box segmentation module is used to perform visual inspection of plastic waste on the non-plastic impurity rejection optimized image and perform bounding box instance segmentation processing to extract a bounding box image of each plastic waste; The detection optimization module is used to perform dynamic component learning and iterative detection optimization on the bounding box image of each plastic waste to build an intelligent waste detection optimization model.

[0007] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the plastic waste detection method described in any one of the above items are implemented.

[0008] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned methods for detecting plastic waste are implemented.

[0009] The beneficial effects of the present invention are as follows: the acquisition of multi-angle images helps to fully understand the situation of the area to be detected and provide more information for subsequent processing; multi-scale detail reconstruction and background feature interference suppression improve image quality and reduce interference, making subsequent processing more accurate and reliable; target visual detection and material component mining help to identify the specific characteristics of each object and provide basic information for subsequent processing; generating material component characteristics of objects helps to further analyze and distinguish different objects and improve the accuracy and reliability of detection; deep semantic segmentation helps to segment objects more finely and improve the accuracy of object recognition; image impurity removal removes interference, making subsequent processing more focused on the main target and improving overall processing efficiency; visual detection and bounding box instance division help to accurately locate and identify plastic waste and provide target objects for subsequent processing; extracting the bounding box image of each plastic waste can better analyze and process the waste; dynamic component learning and iterative detection optimization help to continuously optimize the detection model and improve the accuracy and robustness of waste detection; constructing an intelligent waste detection optimization model to realize the automation and intelligence of waste detection and improve the efficiency and accuracy of waste detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a schematic diagram of the steps of a method for detecting plastic waste according to the present invention; Figure 2 Detailed implementation flow chart of step S1; Figure 3 Detailed implementation flow chart of step S2; Figure 4 Detailed implementation flow chart of step S3. DETAILED DESCRIPTION

[0011] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0012] The present application example provides a method, device, equipment and storage medium for detecting plastic waste. The execution subjects of the method, device, equipment and storage medium for detecting plastic waste include but are not limited to: mechanical equipment, data processing platform, cloud server node, network upload device, etc. equipped with the system can be regarded as the general computing node of the present application, and the data processing platform includes but is not limited to: at least one of an audio image management system, an information management system, and a cloud data management system.

[0013] See also Figures 1 to 4 The present invention provides a method for detecting plastic waste, which comprises the following steps: Step S1: acquiring multi-angle images of the area to be detected; reconstructing multi-scale details of the multi-angle images of the area to be detected, and suppressing interference of regional background features, thereby constructing a background suppression optimized image; Step S2: Performing visual detection of targets within the region of the background suppression optimized image and mining the material composition of the object, thereby generating material composition features of each object; Step S3: Perform deep semantic segmentation based on the material composition characteristics of each object, and perform image impurity removal processing to construct a non-plastic impurity removal optimized image; Step S4: performing visual inspection of plastic waste on the non-plastic impurity rejection optimization image, and performing bounding box instance segmentation processing, thereby extracting a bounding box image of each plastic waste; Step S5: Perform dynamic component learning and iterative detection optimization on the bounding box image of each plastic waste to build an intelligent waste detection optimization model.

[0014] The present invention helps to fully understand the situation of the area to be detected by acquiring multi-angle images, and provides more information for subsequent processing. Multi-scale detail reconstruction and background feature interference suppression improve image quality, reduce interference, and make subsequent processing more accurate and reliable. Target visual detection and material component mining help to identify the specific characteristics of each object, provide basic information for subsequent processing, generate material component characteristics of objects to help further analyze and distinguish different objects, and improve the accuracy and reliability of detection. Deep semantic segmentation helps to segment objects more finely and improve the accuracy of object recognition. Image impurity removal removes interference, so that subsequent processing is more focused on the main target and improves overall processing efficiency. Visual detection and bounding box instance division help to accurately locate and identify plastic waste, provide target objects for subsequent processing, extract the bounding box image of each plastic waste to better analyze and process the waste further, dynamic component learning and iterative detection optimization help to continuously optimize the detection model, improve the accuracy and robustness of waste detection, and build an intelligent waste detection optimization model to realize the automation and intelligence of waste detection, thereby improving the efficiency and accuracy of waste detection.

[0015] In the embodiment of the present invention, refer to Figure 1 , is a schematic flow chart of the steps of a method for detecting plastic waste of the present invention. In this example, the steps of the method include: Step S1: acquiring multi-angle images of the area to be detected; reconstructing multi-scale details of the multi-angle images of the area to be detected, and suppressing interference of regional background features, thereby constructing a background suppression optimized image; In this embodiment, a suitable imaging device is selected, such as a high-resolution camera or a multi-spectral camera, to ensure that high-quality multi-angle images can be obtained. These devices should have good optical performance and be able to work normally under different lighting conditions. The environmental conditions of the area to be detected are determined to ensure uniform light and avoid interference of shadows and reflections on image acquisition. When necessary, artificial lighting or shielding is used to optimize the lighting conditions. The parameters of the camera, such as focal length, shutter speed and ISO value, are set to adapt to different shooting scenes and characteristics of the target object. Images of the area to be detected are captured from different angles and positions. For example, the camera is rotated around the target object, or the pitch angle of the camera is adjusted to ensure that the complete characteristics of the target object are captured from multiple perspectives. Take multiple images at each angle to increase the diversity of data, especially when the surface features of the target object are complex. Improve the quality of the image by taking multiple shots. Classify and store the acquired images according to the shooting angle and time to ensure that they can be quickly accessed and used during subsequent processing. Record the shooting parameters and environmental conditions of each image for subsequent analysis. Preprocess the acquired multi-angle images and apply denoising algorithms (such as Gaussian filtering, median filtering, etc.) to reduce the impact of noise in the image and ensure the accuracy of detail reconstruction. Align multiple images according to the shooting angle and position to ensure that images at different angles are processed in the same coordinate system. This step uses feature matching algorithms (such as SIFT or ORB ) is used to achieve this, and wavelet transform or Laplace pyramid and other technologies are applied to reconstruct the details of multiple aligned images. These methods can extract image detail information at different scales, and retain the details of the target object during the reconstruction process by separating high-frequency information (details) from low-frequency information (background), while suppressing unnecessary background information, and synthesizing detail information of different scales to generate a reconstructed image containing multi-scale details. This process is achieved through weighted averaging or fusion technology to ensure a natural transition of details. Background modeling algorithms (such as Gaussian mixture models) are used to analyze the background features in the reconstructed image, extract the statistical features of the background, and help identify which parts belong to the background and which parts are target objects. The image is divided into different regions (e.g., based on superpixel segmentation technology) to perform interference suppression at the regional level. Each region will be analyzed to determine whether it contains the target object or is interfered by the background. The interference suppression algorithm is applied to the extracted background features. The background information is subtracted from the reconstructed image using background subtraction technology to highlight the foreground target. The image is further cleaned up in combination with morphological processing (such as opening and closing operations) to remove noise and small interference areas and ensure the clarity of the target object. The final image generated should be a background suppression optimized image that can clearly display the target object and reduce the impact of background feature interference on subsequent analysis. The various parameters and results in the processing process are recorded to ensure the traceability of subsequent analysis.

[0016] Step S2: Performing visual detection of targets within the region of the background suppression optimized image and mining the material composition of the object, thereby generating material composition features of each object; In this embodiment, the image to be analyzed is extracted from the generated background suppression optimized image to ensure that the image maintains high quality and clarity for subsequent analysis. The image is preprocessed, such as normalization and enhancement, to improve the accuracy of visual detection. A suitable target detection algorithm is selected, such as YOLO (YouOnlyLookOnce), FasterR-CNN or SSD (Single Shot Multi Box Detector), these algorithms are known for their fast and efficient detection capabilities, and can detect multiple targets in an image at the same time. Set the parameters of the algorithm according to specific needs, such as the size of the input image, confidence threshold, and non-maximum suppression (NMS) parameters to optimize the detection results. Input the background suppression optimized image into the selected target detection model to detect targets in the area. The model will output the bounding box, category, and confidence score of each target, record the detection results, including the location information (such as coordinates and size), category label, and confidence of each object for subsequent analysis, and select the appropriate material component mining method, such as multispectral feature analysis, spectral reflectance measurement, or learning-based methods (such as features extracted by convolutional neural networks). Determine the type of material features that need to be extracted, such as spectral features, texture features, and gloss properties, which can effectively distinguish between different materials. object, process each detected target area, extract the corresponding image data, crop the image of each object according to the bounding box information, ensure that only the target area is analyzed, preprocess the cropped image (such as normalization and enhancement) to improve the effect of subsequent feature extraction, apply the selected material component mining technology to extract the corresponding features from the image of each object, for example, use spectral analysis technology to obtain the spectral curve of the object, or use texture analysis method to extract surface features, record the material component characteristics of each object, including spectral characteristics, texture characteristics and gloss properties, etc. These features will be used as the basis for subsequent material identification and classification, organize the detection information of each object and the corresponding material component characteristics into a structured form, and record them in a database or file for subsequent retrieval and analysis, to ensure that the detection result of each object corresponds to its material characteristics, which is convenient for subsequent research and application.

[0017] Step S3: Perform deep semantic segmentation based on the material composition characteristics of each object, and perform image impurity removal processing to construct a non-plastic impurity removal optimized image; In this embodiment, the acquired background suppression optimized images are used as input. These images should have removed most of the background interference to ensure the clarity of the target object. The image quality and resolution are ensured to be suitable for deep semantic segmentation to ensure the accuracy of subsequent processing. A suitable deep learning semantic segmentation model is selected, such as U-Net, DeepLabv3+ or MaskR-CNN. These models perform well in image segmentation tasks and can effectively identify and separate target objects from backgrounds. According to specific task requirements, the hyperparameters of the model, such as learning rate, batch size and loss function, are set to optimize the training effect. The input images are manually annotated to generate segmentation labels for training. These labels should clearly identify each target object and its boundaries to ensure the accuracy of the training data. The annotated data set is divided into a training set, a validation set and a test set to train and evaluate the model. The selected deep learning model is trained using the annotated data set. The model parameters are continuously optimized through the back propagation algorithm to improve the segmentation accuracy. The loss value and accuracy rate during the training process are recorded to ensure that the model is accurate during the training process. The trained semantic segmentation model converges in the training process, and the training strategy is adjusted according to the performance of the validation set. The trained semantic segmentation model is applied to the background suppression optimization image for deep semantic segmentation. The model will output the category label of each pixel, identify the object and background, and the generated segmentation result will contain the boundary information and category label of each object, providing a basis for the subsequent impurity removal. According to the segmentation results, the non-plastic impurity parts in the image are identified. These impurities include fragments of other materials, dirt or noise. A threshold-based segmentation method or a region-based growth algorithm is used to determine which areas should be regarded as impurities. For the identified non-plastic impurity areas, they are removed. Mask operations are used to delete these areas from the image or replace them with background information. Morphological processing techniques (such as opening and closing operations) are combined to further clean up the removed image to ensure the integrity and clarity of the target object. The image after impurity removal is saved as a non-plastic impurity removal optimization image. This image should clearly show the target object and be free from interference from impurities. The parameters and results of the processing process are recorded to ensure the traceability of the removal process to support subsequent analysis and application.

[0018] Step S4: performing visual inspection of plastic waste on the non-plastic impurity rejection optimization image, and performing bounding box instance segmentation processing, thereby extracting a bounding box image of each plastic waste; In this embodiment, the obtained non-plastic impurity rejection optimized images are used as input. These images should clearly show the features of the target objects and have removed unnecessary background interference. The quality and resolution of the images are ensured to be suitable for visual detection to improve the accuracy of subsequent processing. A suitable target detection algorithm is selected, such as YOLO, FasterR-CNN or SSD. These algorithms can efficiently detect and locate multiple targets in the image. The hyperparameters of the model are adjusted according to specific task requirements, such as the size of the input image, the confidence threshold and the non-maximum suppression (NMS) threshold to optimize the detection effect. The non-plastic impurity rejection optimized images are manually annotated to generate target detection labels for training. These labels should accurately identify the bounding box and category of each plastic waste to ensure the accuracy of the training data. The annotated data set is divided into a training set, a validation set and a test set to facilitate model training and evaluation. The selected target detection model is trained using the annotated data set. The model parameters are continuously optimized through the back propagation algorithm to improve the detection accuracy. The loss value and accuracy rate during the training process are recorded to ensure that the model converges during the training process, and the training strategy is adjusted according to the performance of the validation set. The trained target detection model is applied to the non-plastic impurity rejection optimized image to perform visual inspection of plastic waste. The model will output the bounding box, category label and confidence score of each detected plastic waste, and record the detection results, including the location information (such as coordinates and size), category label and confidence of each plastic waste, for subsequent analysis. According to the bounding box information of the detection results, each detected plastic waste is instanced to ensure that each bounding box accurately contains the corresponding plastic waste and does not overlap. Combined with the non-maximum suppression (NMS) algorithm, the bounding boxes with an overlap higher than the set threshold are removed to ensure the accuracy of the final detection results. The bounding box image of each plastic waste is extracted from the non-plastic impurity rejection optimized image. According to the detected bounding box coordinates, the image of each object is cropped to ensure that only the target area is analyzed. The location information and size of each bounding box image are recorded for subsequent application and analysis. The detection information of each plastic waste and the corresponding bounding box image are organized into a structured form and recorded in a database or file for subsequent retrieval and analysis. It is ensured that the detection result of each plastic waste corresponds to its bounding box image, which is convenient for subsequent research and application.

[0019] Step S5: Perform dynamic component learning and iterative detection optimization on the bounding box image of each plastic waste to build an intelligent waste detection optimization model.

[0020] In this embodiment, a bounding box image of each plastic waste is obtained to ensure that these images cover different types of plastic waste so that the model can learn diverse features. The image is preprocessed, including normalization and enhancement, to improve the learning effect, including random rotation, scaling and color adjustment techniques to increase the diversity of the data. A suitable dynamic component learning algorithm is selected. It is recommended to use transfer learning or incremental learning methods, which can quickly update the model when new data arrives to improve adaptability and real-time performance. Key learning parameters such as learning rate, update frequency and optimization algorithm (such as Adam or SGD) are determined to adapt to dynamic data streams. A pre-trained target detection model (such as YOLO, FasterR-CNN) is selected and fine-tuned based on the collected bounding box images. The fine-tuning makes the model better adapt to specific plastic waste features. The previously labeled bounding box data is used for preliminary training to ensure that the model can effectively identify and distinguish different types of plastic waste. During the model training process, new bounding box image data is continuously introduced. These new data are used for dynamic learning, and the model will update its parameters in real time. To adapt to feature changes in new data, record the model performance after each update, including indicators such as accuracy and recall, to evaluate the learning effect, regularly evaluate the detection performance of the model, use the validation set for testing, calculate various indicators (such as precision, recall and F1-score), ensure the robustness of the model under different conditions, analyze the performance of the model on specific types of plastic waste, identify areas or categories where the model performs poorly, and implement optimization strategies based on the evaluation results, such as adjusting the model's hyperparameters, increasing the diversity of training data, or introducing a more complex model architecture. Use ensemble learning methods to combine the prediction results of multiple models to improve the overall detection performance. Integrate the iteratively optimized models into an intelligent waste detection optimization model to ensure that the model can respond quickly and perform accurate detection in actual applications. Perform final model verification to ensure that it can operate stably in different environments and conditions. Record the model's architecture, training process, performance evaluation and optimization strategy for subsequent maintenance and iterative updates. Update the model regularly to adapt to new data changes and detection needs to ensure its continued effectiveness.

[0021] In this embodiment, refer to Figure 2 , is a flowchart of detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Step S11: Acquire multi-angle images of the area to be detected; Step S12: reconstructing multi-scale details of the multi-angle images of the detection area to construct a multi-scale detail reconstructed image; Step S13: performing environmental background visual analysis on the multi-scale detail reconstructed image to generate environmental background visual features in the image; Step S14: performing multi-angle feature capture based on the environmental background visual features in the image, thereby generating multi-angle environmental background visual features; Step S15: Based on the multi-angle environmental background visual features, the multi-scale detail reconstructed image is subjected to regional background feature interference suppression, thereby constructing a background suppression optimized image.

[0022] In this embodiment, a high-resolution camera or camera is selected to ensure that the device can capture the details of the area to be detected. Multiple cameras are used as needed to achieve multi-angle shooting to improve the comprehensiveness of the image. Camera parameters (such as exposure time, ISO, white balance, etc.) are configured to adapt to different lighting conditions to ensure image quality. A detailed shooting plan is formulated to determine the position and shooting angle of each camera. Generally, all sides of the area to be detected should be covered to ensure that there are no blind spots. On-site inspections are conducted to ensure the stability and safety of the shooting position to avoid obstructions during the shooting process. At the set shooting time, the camera equipment is started to collect multi-angle images in real time to ensure that each camera shoots at the predetermined angle and position. Use synchronization triggers to ensure that all cameras shoot at the same time to ensure the temporal consistency of images during subsequent processing. Convert the acquired multi-angle images into a unified format (such as JPEG, PNG) for subsequent processing. Ensure that the resolution and color mode of the images are consistent. Perform image denoising and enhancement. Use filters (such as Gaussian filtering) to remove image noise to improve the quality of subsequent detail reconstruction. Select appropriate multi-scale reconstruction algorithms, such as wavelet transform, Laplace pyramid or deep learning models (such as convolutional neural networks). These methods can effectively extract and reconstruct image details at different scales. Determine the scale parameters of the reconstruction so that sufficient detail information can be retained during the reconstruction process. Apply the selected Reconstruction algorithm is used to process multi-angle images and generate multi-scale detailed reconstructed images. Through multi-level reconstruction methods, the details and clarity of the images are enhanced. The reconstructed images are fused to form a comprehensive image with rich details, ensuring that information from different angles is effectively integrated. Appropriate visual parsing methods are selected, such as image segmentation, edge detection, or feature extraction models based on deep learning (such as MaskR-CNN). These methods can extract feature information of the environmental background from the image, determine the parameters and models of feature extraction, and facilitate subsequent processing. The multi-scale detailed reconstructed images are subjected to visual parsing of the environmental background, and the main background features in the image (such as color, texture) are extracted. , shape, etc.), generate a feature map of the environmental background, record the spatial position and importance of each feature for subsequent analysis and application, select appropriate feature matching and fusion technology, such as SIFT (scale-invariant feature transform), SURF (speeded up robust features) or ORB, to extract multi-angle features of the environmental background, determine the parameters of feature matching to adapt to the changes of features at different angles, and capture multi-angle features of the image based on the extracted visual features of the environmental background. Through the feature matching algorithm, compare images at different angles to extract common background features, generate a multi-angle environmental background visual feature data set, record the performance of each feature at different angles and its spatial distribution, and select appropriate Interference suppression algorithms, such as background modeling (such as Gaussian mixture model), image fusion technology or deep learning models (such as autoencoders) are used to suppress background interference. The suppression parameters and algorithms are determined to adapt to different types of interference. Multi-angle environmental background visual features are input into the suppression algorithm, and interference suppression of regional background features is performed. By comparing background features with interference features, unnecessary interference information is removed, and background suppression optimized images are generated to ensure that important target features are retained while effectively eliminating the impact of background interference. The generated background suppression optimized images are stored in the database to ensure the traceability and integrity of the images, and the key parameters in the processing process are recorded for subsequent analysis and improvement. The optimization results are evaluated and the suppression effect is analyzed to provide a reliable basis for subsequent detection and application.

[0023] In this embodiment, the specific steps of step S12 are: Performing multi-region proportional segmentation on the multi-angle images of the detection area to obtain multiple image areas; Calculating the local contrast between multiple regions of the image, thereby obtaining the local contrast between multiple regions; Adaptively enhancing contrast according to local contrasts among multiple regions, thereby generating a contrast-enhanced image; Performing brightness distribution analysis on the contrast-enhanced image to generate image brightness distribution data; Perform nonlinear brightness equalization based on image brightness distribution data to construct a brightness-equalized enhanced image; Perform discrete wavelet transform on the brightness equalization enhanced image to obtain multiple images of different scales; Calculate the frequency components of multiple images of different scales one by one to generate frequency component features of each scale; According to the frequency component characteristics of each scale, the edge details of the corresponding scale are sharpened, thereby generating edge detail sharpening features of each scale; Based on the edge detail sharpening features of each scale, multi-scale feature fusion and reconstruction are performed on multiple images of different scales to construct a multi-scale detail reconstructed image.

[0024] In this embodiment, an image to be processed is obtained from a multi-angle image set to ensure the clarity and resolution of the image for subsequent processing. The image is preprocessed, such as denoising and enhancement, to ensure that the image quality meets the segmentation requirements. A uniform grid segmentation algorithm is used to divide the image into multiple small areas. An image processing library such as OpenCV is used to set the same width and height for segmentation, determine the segmentation ratio and the number of areas, and adjust according to the characteristics of the image and analysis requirements. The image of the detection area is processed using the selected segmentation algorithm to generate multiple geometrically segmented image areas. The size and number of each area should be within a preset range. The location information and features of each area are recorded for subsequent processing and analysis. A local contrast calculation method is selected, such as local contrast enhancement. Enhancement) or standard deviation calculation, evaluate contrast by calculating the degree of change of pixel values ​​in each region, determine the calculation window size to facilitate the analysis of contrast changes in local regions, calculate local contrast for each segmented image region, generate contrast values ​​between multiple regions, use the mean and standard deviation of pixel intensity to calculate contrast, record the calculation results in a data structure to facilitate subsequent contrast enhancement processing, use methods such as adaptive histogram equalization (CLAHE) or gamma correction to perform contrast enhancement to facilitate adaptive contrast adjustment in different regions, determine enhancement parameters such as block size and contrast limit to optimize the enhancement effect, apply the calculated local contrast value to the adaptive contrast enhancement algorithm, generate a contrast-enhanced image, ensure that image details are preserved during the enhancement process, record the parameters and effects of the enhanced image for subsequent analysis and optimization, use the histogram method to calculate the brightness distribution of the image, record the number of pixels with different brightness values, and generate brightness distribution data, use OpenCV or MATLAB Use tools such as , and analyze its shape and features. Analyze the brightness distribution of the contrast-enhanced image and generate brightness distribution data, including the mean and variance of the brightness value. Record the analysis results for subsequent nonlinear brightness equalization processing. Select a nonlinear brightness equalization algorithm, such as adaptive histogram equalization, or use gamma correction to achieve brightness equalization. Determine the equalization parameters to optimize the brightness distribution of the image. Input the brightness distribution data into the nonlinear equalization algorithm. Process the contrast-enhanced image to generate a brightness-equalized enhanced image. Record the parameters and effects of the equalization processing and visualize them to ensure that users can intuitively understand the processing results. Select the discrete wavelet transform (DWT) method. Decompose the image into frequency components of different scales through the wavelet function. Select a suitable wavelet basis (such as Haar, Daubechies) and the number of decomposition layers to ensure the processing effect. Perform discrete wavelet transform on the brightness-equalized enhanced image to generate multiple images of different scales. Record the details and approximate components of each scale. Save the transformation results for subsequent frequency component calculation and feature extraction.Select the frequency component calculation method, such as the amplitude calculation after Fourier transform or wavelet transform, to extract the frequency features of each scale, determine the calculation parameters to optimize the extraction effect of frequency features, calculate the frequency components of the wavelet transform results of each scale one by one, generate the frequency component features of each scale, record the calculated features in the data structure to facilitate subsequent sharpening processing, select the appropriate edge detail sharpening method, such as Laplace operator, Sobel operator or high-pass filter, to enhance the edge details of the image, determine the sharpening parameters, such as the filter size and enhancement factor, to ensure the processing effect, based on each scale The frequency component features of the image are used to sharpen the edge details of the corresponding scale, generate the edge detail sharpening features of each scale, record the parameters and effects of the sharpening results for subsequent analysis and display, select multi-scale feature fusion methods, such as weighted average, maximum pooling or deep learning methods for feature fusion, to generate a comprehensive multi-scale reconstructed image, determine the fusion strategy and weights to adapt to the importance of features at different scales, and perform multi-scale feature fusion reconstruction on multiple images of different scales based on the edge detail sharpening features of each scale to generate the final multi-scale detail reconstructed image, ensuring that the image is rich in details and has good visual effects.

[0025] In this embodiment, refer to Figure 3 , is a flowchart of detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Step S21: performing visual detection of targets within the region on the background suppression optimized image, and marking object nodes within multiple regions; Step S22: performing multispectral feature mining on target nodes in multiple regions to generate spectral features for each object; Step S23: Analyze the surface texture characteristics of each object's spectral characteristics to obtain the surface texture characteristics of each object; Step S24: performing visual gloss attribute analysis on the spectral features of each object to extract the visual gloss attribute of each object; Step S25: mining the material components of each object based on the visual gloss properties of each object and the surface texture features of each object, thereby generating material component features of each object.

[0026] In this embodiment, the area to be detected is extracted from the background suppression optimized image to ensure the quality and clarity of the image for subsequent target detection. The image is preprocessed, such as denoising and enhancement, to improve the accuracy of target detection. A suitable target detection algorithm is selected, such as YOLO, Faster R-CNN or SSD. These algorithms can quickly and accurately detect multiple targets in the image. The configuration parameters of the algorithm, such as the input image size, detection threshold, etc., are determined to optimize the detection effect. The background suppression optimized image is input into the selected target detection model to detect objects in the area. The model will identify each target and generate its bounding box, mark the detected object nodes, record the location information (such as coordinates and size) of each object, generate target detection result data, select multispectral imaging technology, acquire spectral information of different bands by performing multispectral image acquisition on each target node, and use a multispectral camera or sensor to perform data Acquisition, determine the band range that needs to be collected to cover the characteristic spectrum of the target material, extract multi-spectral features for each marked target node, generate spectral feature data for each object by analyzing the reflectivity of different bands, and record the spectral data in the database, including band information, reflectivity value, etc., to facilitate subsequent analysis and comparison, use gray level co-occurrence matrix (GLCM), local binary pattern (LBP) or wavelet transform and other methods to perform surface texture analysis, extract the texture features of the object surface by analyzing the details of the texture, and determine the analysis parameters, such as window size and number of features, to ensure the comprehensiveness and accuracy of the analysis, for each object Spectral feature data is used to analyze surface texture features, and feature data related to the surface texture of the object (such as contrast, entropy, uniformity, etc.) is generated. The extracted texture features are recorded in the data structure for subsequent material component mining. Visual gloss attribute analysis methods are selected, such as reflectance analysis, gloss measurement, or gloss feature extraction models based on deep learning. These methods can accurately evaluate the gloss characteristics of the object surface, determine the analysis parameters and models to adapt to the gloss characteristics of different materials, perform visual gloss attribute analysis on the spectral features of each object, extract related attribute features such as gloss and reflectance, and record gloss attribute data, including the surface of the object. Reflection characteristics and glossiness indicators are used for subsequent material composition analysis. Appropriate material composition analysis methods are selected, such as support vector machines (SVM), random forests, or classification models based on deep learning. These models can identify the composition of materials based on spectral and texture features. The input features of the model, including spectral features, texture features, and gloss properties, are determined to facilitate accurate classification. The visual gloss properties and surface texture features of each object are input into the material composition analysis model to identify and classify the material of the object. Material composition feature data for each object is generated, including the identified material type and related feature values, and recorded in the database for subsequent use.

[0027] In this embodiment, refer to Figure 4 , is a flowchart of detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: Step S31: performing non-plastic impurity feature recognition on object nodes in multiple regions based on the material composition features of each object, and extracting multiple non-plastic impurities; Step S32: performing deep semantic segmentation on the background suppression optimized image to obtain multiple object type region images; Step S33: performing image impurity removal processing on the regional images of multiple object types according to the multiple non-plastic impurities, and constructing a non-plastic impurity removal optimized image.

[0028] In this embodiment, the material composition features of each object extracted in the previous step are collected and sorted, including spectral features, texture features, and gloss properties. These features will be used to distinguish plastic from non-plastic impurities, ensure the accuracy and consistency of the data, and facilitate subsequent impurity identification. Select a suitable machine learning algorithm, such as random forest, support vector machine (SVM) or deep learning model (such as convolutional neural network) to identify non-plastic impurities. These algorithms can classify based on material features, determine the hyperparameter configuration and training strategy of the recognition model, so that the model can effectively distinguish different materials, and input the collected material composition features into the selected recognition model. In the model, feature recognition of non-plastic impurities is performed. The model will output the impurity type and its confidence for each object, record the recognition results, including the type, quantity and location of the identified non-plastic impurities in the image, and form an impurity feature report. Use background suppression to optimize the image as input to ensure the quality and clarity of the image for semantic segmentation. Select a suitable deep learning semantic segmentation model, such as U-Net, DeepLab or MaskR-CNN. These models can effectively identify and segment different object categories in the image. If a pre-trained model is used, fine-tune it according to the dataset; if training from scratch, it is necessary Prepare annotated training data sets to ensure that each object category has enough samples in the image, determine training parameters such as learning rate, batch size, and number of training rounds to optimize model performance, input the background suppression optimized image into the deep semantic segmentation model, perform segmentation operations, and generate multiple object category region images. The model will output the category label of each pixel and record the segmentation results, including the region boundaries and pixel information of each object category for subsequent processing and analysis. According to the identified non-plastic impurities, formulate corresponding image processing strategies, select pixel-based mask processing or region removal methods, and determine the removal conditions, such as the size and shape of the impurities. The shape and position of the impurities are identified to ensure the effectiveness of the removal process. According to the identified non-plastic impurity information, the object type area image generated after deep semantic segmentation is processed. The image processing library (such as OpenCV) is used to remove impurities and generate a non-plastic impurity removal optimized image. The removed area is filled or smoothed to ensure the naturalness and integrity of the image. The generated non-plastic impurity removal optimized image is saved in the database to ensure the traceability of the processing results. The key parameters and effects of the removal process are recorded, the removal results are evaluated, and the effectiveness of the removal and the impact on the overall image quality are analyzed to provide a basis for subsequent optimization.

[0029] In this embodiment, step S4 includes the following steps: Step S41: Perform visual inspection of plastic waste on the non-plastic impurity rejection optimization image and mark each plastic waste node; Step S42: performing regional precise positioning calculation on each plastic waste node to obtain the positioning coordinates of each plastic waste node; Step S43: performing three-dimensional morphological analysis of each plastic waste node based on the positioning coordinates of each plastic waste node, and generating three-dimensional morphological features of each plastic waste node; Step S44: performing bounding box instance division processing on the non-plastic impurity elimination optimized image according to the three-dimensional morphological features of each plastic waste, thereby extracting the bounding box image of each plastic waste.

[0030] In this embodiment, the image to be processed is extracted from the non-plastic impurity rejection optimized image to ensure its quality and clarity for subsequent plastic waste detection. The image is denoised and preprocessed to improve the accuracy of subsequent detection. A suitable target detection algorithm is selected, such as YOLO (YouOnlyLookOnce), FasterR-CNN or EfficientDet, which effectively identifies and marks plastic waste nodes. The hyperparameters of the algorithm, such as detection threshold and confidence, are set to optimize the detection effect. The image after rejection optimization is input into the selected target detection model for visual detection of plastic waste. The model will output the bounding box and category label of each detected plastic waste, record the detection results, including the location information (such as coordinates and size) of each plastic waste node, and mark it on the image to form a detection result image. It will select an appropriate positioning algorithm, such as center point positioning or bounding box calculation, to ensure that the coordinate information of each plastic waste node can be accurately obtained, and determine the calculation parameters, such as positioning accuracy and coordinate system settings, for subsequent analysis. It will perform regional precise positioning calculations for each detected plastic waste node, calculate the center coordinates of each node based on the bounding box information in the detection results, and record the positioning results in a data structure, including The precise coordinates (x, y) and corresponding bounding box size of each waste node are obtained for subsequent processing. Appropriate 3D morphological analysis methods are selected, such as point cloud-based analysis, 3D reconstruction technology or shape descriptor extraction (such as Hu moment). These methods can effectively analyze the 3D characteristics of plastic waste and determine the 3D characteristic indicators that need to be analyzed, such as volume, surface area, contour characteristics, etc. Based on the positioning coordinates of each plastic waste node, the 3D point cloud data or image information of the corresponding area is extracted to perform 3D morphological analysis of each waste item one by one, and the 3D morphological characteristic data of each plastic waste is generated, including volume, surface shape, contour and other geometric features, and recorded. Record them in the database, select a suitable bounding box division algorithm, commonly used methods include pixel-based mask processing or region growing algorithm, ensure that accurate bounding boxes can be extracted according to the three-dimensional morphological features, determine the division standards and parameters to ensure that different plastic waste instances can be effectively distinguished, and perform bounding box instance division processing on the optimized image after non-plastic impurities are removed according to the three-dimensional morphological characteristics of each plastic waste. Use the previously calculated positioning coordinates and three-dimensional feature data to accurately extract the bounding box image of each plastic waste, generate the final bounding box image of each plastic waste, and record the position information and size of the bounding box for subsequent application and analysis.

[0031] In this embodiment, step S5 includes the following steps: Step S51: performing dynamic component learning on each bounding box image of plastic waste, thereby generating plastic waste detection learning data; Step S52: Calculating the detection accuracy of the plastic waste detection learning data, thereby generating a waste detection accuracy parameter; Step S53: Perform iterative detection optimization based on the waste detection accuracy parameters to build an intelligent waste detection optimization model.

[0032] In this embodiment, the bounding box image of each plastic waste is collected to ensure that the image quality is good to facilitate subsequent dynamic component learning. These images should contain plastic waste at different angles and under different background conditions. The images are properly preprocessed, such as normalization and enhancement, to improve the effect of the learning process. A suitable dynamic learning algorithm is selected, such as transfer learning, online learning, or incremental learning. These methods quickly update the model when new data arrives and continuously improve the detection performance. The learning parameter settings, such as learning rate, batch size, and update frequency, are determined to adapt to the dynamic data flow. The bounding box image of each plastic waste is input into the dynamic learning model for feature extraction and learning. The model will update its internal weights according to the input data and gradually improve its recognition ability for plastic waste. The loss function and accuracy rate during the learning process are recorded to ensure that the model is continuously optimized during the learning process. The generated learning data will be used for subsequent detection tasks. A suitable detection accuracy calculation method is selected, such as precision and recall. ll) and F1-score, these indicators can comprehensively reflect the detection ability of the model, determine the evaluation thresholds and standards, so as to effectively classify and evaluate the detection results, use plastic waste detection learning data to evaluate the detection results of the model, calculate various accuracy indicators according to the real labels and prediction results, record the evaluation results, including the value of each indicator, and the performance of the model under different conditions, to provide a basis for subsequent optimization, select appropriate model optimization methods, such as hyperparameter adjustment, model integration or use of more complex network architectures, the goal of optimization is to improve detection accuracy and robustness, determine the optimization strategy and process, so as to facilitate systematic iterative optimization under the guidance of detection accuracy parameters, based on waste detection accuracy parameters, perform iterative detection optimization, adjust the model structure or training process for aspects with poor performance in the evaluation results, conduct a new round of training and evaluation, record the optimized detection accuracy parameters, continuously improve the model performance through multiple iterations, and finally build an intelligent waste detection optimization model.

[0033] In this embodiment, the present invention further provides a plastic waste detection device, comprising: The image optimization module is used to obtain multi-angle images of the area to be detected; reconstruct multi-scale details of the multi-angle images of the area to be detected, and suppress the interference of regional background features, so as to construct a background suppression optimized image; The material composition recognition module is used to perform visual detection of targets within the background suppression optimized image and to mine the material composition of the object, thereby generating the material composition features of each object; The impurity removal module is used to perform deep semantic segmentation based on the material composition characteristics of each object, and to remove impurities from the image to construct an optimized image for non-plastic impurity removal; A bounding box segmentation module is used to perform visual inspection of plastic waste on the non-plastic impurity rejection optimized image and perform bounding box instance segmentation processing to extract a bounding box image of each plastic waste; The detection optimization module is used to perform dynamic component learning and iterative detection optimization on the bounding box image of each plastic waste to build an intelligent waste detection optimization model.

[0034] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method for detecting plastic waste described in any one of the above items is implemented.

[0035] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method for detecting plastic waste described in any one of the above items are implemented.

[0036] Those skilled in the art clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0037] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it is stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or partly or all or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media for storing program codes such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks or optical disks.

[0038] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0039] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for detecting plastic waste, characterized in that: The following steps are involved: Step S1: acquiring multi-angle images of the area to be detected; reconstructing multi-scale details of the multi-angle images of the area to be detected, and suppressing interference of regional background features, thereby constructing a background suppression optimized image; Step S2: Performing visual detection of targets within the region of the background suppression optimized image and mining the material composition of the object, thereby generating material composition features of each object; Step S3: Perform deep semantic segmentation based on the material composition characteristics of each object, and perform image impurity removal processing to construct a non-plastic impurity removal optimized image; Step S4: performing visual inspection of plastic waste on the non-plastic impurity rejection optimization image, and performing bounding box instance segmentation processing, thereby extracting a bounding box image of each plastic waste; Step S5: Perform dynamic component learning and iterative detection optimization on the bounding box image of each plastic waste to build an intelligent waste detection optimization model.

2. The method for detecting plastic waste according to claim 1, characterized in that: The specific steps of step S1 are: Step S11: Acquire multi-angle images of the area to be detected; Step S12: reconstructing multi-scale details of the multi-angle images of the detection area to construct a multi-scale detail reconstructed image; Step S13: performing environmental background visual analysis on the multi-scale detail reconstructed image to generate environmental background visual features in the image; Step S14: performing multi-angle feature capture based on the environmental background visual features in the image, thereby generating multi-angle environmental background visual features; Step S15: Based on the multi-angle environmental background visual features, the multi-scale detail reconstructed image is subjected to regional background feature interference suppression, thereby constructing a background suppression optimized image.

3. The method for detecting plastic waste according to claim 2, characterized in that: The specific steps of step S12 are: Performing multi-region proportional segmentation on the multi-angle images of the detection area to obtain multiple image areas; Calculating the local contrast between multiple regions of the image, thereby obtaining the local contrast between multiple regions; Adaptively enhancing contrast according to local contrasts among multiple regions, thereby generating a contrast-enhanced image; Performing brightness distribution analysis on the contrast-enhanced image to generate image brightness distribution data; Perform nonlinear brightness equalization based on image brightness distribution data to construct a brightness-equalized enhanced image; Perform discrete wavelet transform on the brightness equalization enhanced image to obtain multiple images of different scales; Calculate the frequency components of multiple images of different scales one by one to generate frequency component features of each scale; According to the frequency component characteristics of each scale, the edge details of the corresponding scale are sharpened, thereby generating edge detail sharpening features of each scale; Based on the edge detail sharpening features of each scale, multi-scale feature fusion and reconstruction are performed on multiple images of different scales to construct a multi-scale detail reconstructed image.

4. The method for detecting plastic waste according to claim 1, characterized in that: The specific steps of step S2 are: Step S21: performing visual detection of targets within the region on the background suppression optimized image, and marking object nodes within multiple regions; Step S22: performing multispectral feature mining on target nodes in multiple regions to generate spectral features for each object; Step S23: Analyze the surface texture characteristics of each object's spectral characteristics to obtain the surface texture characteristics of each object; Step S24: performing visual gloss attribute analysis on the spectral features of each object to extract the visual gloss attribute of each object; Step S25: mining the material components of each object based on the visual gloss properties of each object and the surface texture features of each object, thereby generating material component features of each object.

5. The method for detecting plastic waste according to claim 1, characterized in that: The specific steps of step S3 are: Step S31: performing non-plastic impurity feature recognition on object nodes in multiple regions based on the material composition features of each object, and extracting multiple non-plastic impurities; Step S32: performing deep semantic segmentation on the background suppression optimized image to obtain multiple object type region images; Step S33: performing image impurity removal processing on the images of the regions of the multiple object types according to the multiple non-plastic impurities, and constructing a non-plastic impurity removal optimized image.

6. The method for detecting plastic waste according to claim 1, characterized in that: The specific steps of step S4 are: Step S41: Perform visual inspection of plastic waste on the non-plastic impurity rejection optimization image and mark each plastic waste node; Step S42: performing regional precise positioning calculation on each plastic waste node to obtain the positioning coordinates of each plastic waste node; Step S43: performing three-dimensional morphological analysis of each plastic waste node based on the positioning coordinates of each plastic waste node, and generating three-dimensional morphological features of each plastic waste node; Step S44: performing bounding box instance division processing on the non-plastic impurity elimination optimized image according to the three-dimensional morphological features of each plastic waste, thereby extracting the bounding box image of each plastic waste.

7. The method for detecting plastic waste according to claim 1, characterized in that: The specific steps of step S5 are: Step S51: performing dynamic component learning on each bounding box image of plastic waste, thereby generating plastic waste detection learning data; Step S52: Calculating the detection accuracy of the plastic waste detection learning data, thereby generating a waste detection accuracy parameter; Step S53: Perform iterative detection optimization based on the waste detection accuracy parameters to build an intelligent waste detection optimization model.

8. A device for detecting plastic waste, characterized in that: The method for detecting plastic waste according to claim 1 comprises: The image optimization module is used to obtain multi-angle images of the area to be detected; reconstruct multi-scale details of the multi-angle images of the area to be detected, and suppress the interference of regional background features, so as to construct a background suppression optimized image; The material composition recognition module is used to perform visual detection of targets within the background suppression optimized image and to mine the material composition of the object, thereby generating the material composition features of each object; The impurity removal module is used to perform deep semantic segmentation based on the material composition characteristics of each object, and to remove impurities from the image to construct an optimized image for non-plastic impurity removal; A bounding box segmentation module is used to perform visual inspection of plastic waste on the non-plastic impurity rejection optimized image and perform bounding box instance segmentation processing to extract a bounding box image of each plastic waste; The detection optimization module is used to perform dynamic component learning and iterative detection optimization on the bounding box image of each plastic waste to build an intelligent waste detection optimization model.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method for detecting plastic waste according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for detecting plastic waste according to any one of claims 1 to 7 are implemented.

Images