A method for monitoring the purity of plastic recycled materials based on artificial intelligence

Abstract

The application relates to an artificial intelligence-based plastic recycling material purity monitoring method, which comprises the following steps: collecting surface visual image data and spectrum signal data of plastic recycling material, as well as environmental parameter data and corresponding plastic recycling material purity data, and performing data processing; constructing a double-branch deep neural network model, extracting surface visual image feature vectors and spectrum signal feature vectors from the surface visual image data and the spectrum signal data respectively and performing fusion; combining the plastic recycling material fusion feature vectors, training and optimizing the double-branch deep neural network model by using a plastic feature data set; collecting real-time surface visual image data and real-time spectrum signal data of the plastic recycling material to be monitored, as well as real-time environmental parameter data, and inputting the real-time surface visual image data, the real-time spectrum signal data and the real-time environmental parameter data into the plastic recycling material purity monitoring model to obtain a real-time purity value of the plastic recycling material. The application can improve the accuracy and efficiency of monitoring the purity of the plastic recycling material, and solves the problems of low efficiency and large error of artificial detection.

Description

Technical Field

[0001] This application relates to the field of materials testing technology, specifically to an artificial intelligence-based method for monitoring the purity of recycled plastic materials. Background Technology

[0002] The purity monitoring of recycled plastic materials refers to the process of detecting the proportion of target plastic components, the types and contents of impurities in recycled plastics. Its purpose is to assess the quality of recycled materials and determine whether they meet the standards for reuse. This is of great significance for ensuring the quality of plastic products and the sustainable development of the plastic recycling industry.

[0003] In current technologies for monitoring the purity of recycled plastics, traditional near-infrared spectroscopy struggles to identify black plastics due to their strong absorption of near-infrared light. While mid-wave infrared spectroscopy can address this issue, its high equipment cost and reliance on specialized hardware limit its widespread application. Furthermore, spectroscopic techniques lack sufficient sensitivity for detecting low-proportion mixed materials, making it difficult to accurately distinguish plastics with similar chemical compositions. The computer vision technology used is significantly constrained by environmental factors; surface contamination, wear, and changes in lighting can severely impact the accuracy of visual features, causing the recognition model to struggle to operate stably in complex environments. Moreover, traditional visual methods rely solely on appearance features, failing to effectively differentiate plastics with similar chemical compositions and prone to misjudgment. Summary of the Invention

[0004] To address the problems existing in the prior art, this application aims to provide an artificial intelligence-based method for monitoring the purity of recycled plastics. This AI-based method improves the accuracy and efficiency of monitoring the purity of recycled plastics, solving the problems of low efficiency and large errors associated with manual testing.

[0005] The method for monitoring the purity of recycled plastic materials based on artificial intelligence as described in this application includes the following steps:

[0006] S1. Obtain multiple batches of recycled plastic materials with monitored purity in history, and collect surface visual image data and spectral signal data of the recycled plastic materials, as well as environmental parameter data and corresponding recycled plastic material purity data. Perform data processing to obtain a plastic feature dataset.

[0007] S2. Construct a dual-branch deep neural network model, extract surface visual image feature vectors and spectral signal feature vectors from the surface visual image data and the spectral signal data respectively, and fuse them to obtain the fused feature vector of the plastic recycled material;

[0008] S3. Combining the fused feature vector of the plastic recycled material, the dual-branch deep neural network model is trained and optimized using the plastic feature dataset to obtain the plastic recycled material purity monitoring model.

[0009] S4. Collect real-time surface visual image data, real-time spectral signal data, and real-time environmental parameter data of the plastic recycled material to be monitored, and input them into the plastic recycled material purity monitoring model to obtain the real-time purity value of the plastic recycled material and generate the plastic recycled material purity monitoring result.

[0010] Preferably, step S1 specifically includes:

[0011] The surface visual image data includes a surface image of recycled plastic and plastic color information. The plastic color information is the hue, saturation, and brightness information of the corresponding HSV color space obtained based on the RGB color value of each pixel in the surface image of recycled plastic.

[0012] The spectral signal data includes near-infrared spectral data of recycled plastic materials in the wavelength range of 780-2526 nm and 200-3500 cm⁻¹. -1 Raman spectral data in the wavenumber range;

[0013] The environmental parameter data includes temperature, humidity, and light intensity data of the environment in which the recycled plastic material is located;

[0014] The collected surface visual image data, spectral signal data, and environmental parameter data are transmitted to the edge computing gateway.

[0015] Preferably, step S1 further includes:

[0016] The surface image of the recycled plastic material is obtained from the edge computing gateway, and then subjected to median filtering for noise reduction and histogram equalization. The surface image of the recycled plastic material is converted into a grayscale image and the grayscale co-occurrence matrix is ​​calculated.

[0017] The texture pattern of the surface image of the recycled plastic material is extracted using a local binary mode to obtain the texture features of the surface of the recycled plastic material;

[0018] The contour of the plastic recycled material is obtained by edge detection algorithm, and the perimeter, area, roundness and rectangularity of the contour are calculated. The shape features are extracted by Fourier descriptor.

[0019] The surface image of the recycled plastic is detected and analyzed to obtain the scratch length and width, crack length and direction, hole area and diameter, and number and volume of bubbles on the surface of the recycled plastic.

[0020] The near-infrared spectral data were filtered, smoothed, and subjected to principal component analysis for dimensionality reduction to obtain absorbance in the wavelength range of 780-2526 nm, as well as absorbance peaks at wavelengths of 1100 nm, 1450 nm, and 1900 nm.

[0021] Obtain Raman spectra from 200-3500 cm⁻¹ -1 Scattered light intensity within the wavenumber range, and polyethylene 2840cm -1 Polypropylene 800-850cm -1 Polyvinyl chloride 600-700cm -1 The intensity and position of characteristic peaks;

[0022] By combining the temperature data, humidity data, and light intensity data, the processed surface visual image data and spectral signal data are calibrated. The calibrated surface visual image data, spectral signal data, environmental parameter data, and plastic recycling purity data are then correlated, labeled, and integrated to form a plastic feature dataset.

[0023] Preferably, step S2 specifically includes:

[0024] A dual-branch deep neural network model is constructed, comprising a surface visual image data processing branch and a spectral signal data processing branch. The surface visual image data processing branch consists of multiple convolutional layers, pooling layers, and fully connected layers, while the spectral signal data processing branch consists of multiple one-dimensional convolutional layers, global average pooling layers, and fully connected layers.

[0025] The surface visual image data in the plastic feature dataset is input into the surface visual image data processing branch, and then passed through the convolutional layer, pooling layer and fully connected layer in sequence to obtain the surface visual image feature vector;

[0026] The spectral signal data in the plastic feature dataset is input into the spectral signal data processing branch, and then passed through a one-dimensional convolutional layer, a global average pooling layer, and a fully connected layer in sequence to obtain the spectral signal feature vector.

[0027] The surface visual image feature vector and the spectral signal feature vector are fused together to form a plastic recycled material fused feature vector.

[0028] Preferably, step S3 specifically includes:

[0029] The plastic feature dataset is divided into training and testing sets according to the ratio A1:A2, and the parameters of each layer of the dual-branch deep neural network model are initialized.

[0030] Set the training rounds Q, combine the fused feature vector of the plastic recycled material, and use the training set to train the dual-branch deep neural network model for Q iterations;

[0031] The test set is input into the trained dual-branch deep neural network model for testing. The fused feature vector of the plastic recycled material is combined with forward propagation to obtain the predicted purity value of plastic recycled material for all samples in the test set.

[0032] The dual-branch deep neural network model is evaluated based on the predicted purity value of the recycled plastic material.

[0033] Preferably, the model evaluation includes:

[0034] The purity prediction value of the recycled plastic includes a first purity prediction value P based solely on surface visual image features. v The second purity prediction value P based solely on spectral signal characteristics s And the third purity prediction value P of the fusion feature. f Calculate the mean intermodal consistency score C. J and weighted evaluation index M C :

[0035]

[0036] Among them, C i P represents the intermodal consistency score of the i-th sample in the test set; v (i) P represents the first purity prediction value of the i-th sample in the test set; s (i) P represents the predicted second purity value of the i-th sample in the test set; f (i) y represents the predicted third purity value of the i-th sample in the test set; (i) The value represents the actual purity of the i-th sample in the test set; N represents the total number of samples in the test set; i = 1, 2, ..., N;

[0037] Based on the intermodal consistency score C i The sample prediction variance of the test set is used to dynamically shrink the confidence interval, and the confidence interval is calculated:

[0038]

[0039] Where Z represents the dynamic adjustment coefficient for the dynamically shrinking confidence interval; Z0 represents the standard confidence level quantile; and γ represents the adaptive coefficient. This represents the prediction variance of the i-th sample in the test set calculated using Monte Carlo methods; max(σ) 2 CI represents the maximum variance of all samples in the test set; i This represents the confidence interval for the i-th sample in the test set;

[0040] If M C ≤0.5, C J ≥0.7 and CI i If the proportion of samples not covering the true value is ≤20%, then the dual-branch deep neural network model is deemed to meet the requirements, and the dual-branch deep neural network model is used as the purity monitoring model for plastic recycled materials.

[0041] If M C ≤0.5, C J ≥0.7 and CI i If any one of the following conditions is not met: the proportion of samples not covering the true value is ≤20%, then the dual-branch deep neural network model is deemed unqualified, and model optimization is performed until M... C ≤0.5, C J ≥0.7 and CI i The requirement is met if the proportion of samples not covering the true value is ≤20%.

[0042] Preferably, the model optimization includes:

[0043] Based on the obtained weighted evaluation index M C The mean of the intermodal consistency score C J Confidence interval CI i The dual-branch deep neural network model is analyzed and optimized.

[0044] When M C If the value is greater than 0.5, it is determined that feature extraction is insufficient, and a feature extraction optimization strategy is executed until M is satisfied. C ≤0.5;

[0045] When C J If the value is less than 0.7, the feature fusion mechanism is deemed to have a defect, and an optimization strategy for the feature fusion mechanism is executed until C is satisfied. J ≥0.7;

[0046] When CI i If the proportion of samples not covering the true value is greater than 20%, then the confidence interval is determined to be unable to effectively contain the true value, and a confidence interval optimization strategy is executed until the CI is met. i The proportion of samples that do not cover the true value is ≤20%.

[0047] Preferably, step S4 specifically includes:

[0048] Real-time surface visual image data, real-time spectral signal data, and real-time environmental parameter data of the plastic recycled material to be monitored are collected, processed, and then input into the plastic recycled material purity monitoring model to obtain the preliminary purity value P of the plastic recycled material. pre ;

[0049] Based on a pre-established calibration sample library, a similarity matching algorithm is used to retrieve multiple historical samples that are most similar to the characteristics of the current recycled plastic material, and the mean calibration purity value P is calculated. J And the preliminary purity value P pre The real-time purity value P of the recycled plastic material was obtained through calibration. S :

[0050] P S =P pre +α*(P J -P pre )

[0051] Wherein, α represents the adaptive weight, which is dynamically adjusted according to the sample matching degree, and its value ranges from 0 to 1;

[0052] According to the real-time purity value P S The purity of recycled plastic materials is analyzed, and the purity monitoring results of recycled plastic materials are generated.

[0053] Preferably, the purity analysis includes:

[0054] If P S If the purity is ≥95%, the plastic recycling material is determined to be of Grade 1 purity, and the plastic recycling material will continue to undergo the original processing flow.

[0055] If 90% ≤ P S If the purity is less than 95%, the plastic recycling material is determined to be of grade two, an alarm is issued, and the plastic recycling material is subjected to a second re-inspection.

[0056] If 80% ≤ P S If the purity is less than 90%, the plastic recycling material is determined to be of grade three, an alarm is issued, and the plastic recycling material is sorted to the downgrade processing area.

[0057] If P S If the purity is less than 80%, the plastic recycling material is determined to be of grade four, and an alarm message is issued, requiring the plastic recycling material to undergo enhanced impurity removal treatment.

[0058] Preferably, the purity monitoring results of the recycled plastic material include:

[0059] The real-time surface visual image data, real-time spectral signal data, real-time environmental parameter data, and real-time purity value of the same batch of recycled plastic materials are integrated and processed, and a plastic recycled material purity monitoring result containing real-time purity value, purity level, and purity fluctuation trend is generated through visualization technology.

[0060] The method for monitoring the purity of recycled plastic materials based on artificial intelligence described in this application has the following advantages:

[0061] This application presents an AI-based method for monitoring the purity of recycled plastic materials. By simultaneously acquiring surface visual image data and spectral signal data, and combining them with environmental parameter data, a dual-branch deep neural network is used to achieve multi-dimensional feature fusion. This avoids the limitations of single-modality analysis and significantly improves the accuracy and anti-interference capability of purity monitoring. Introducing environmental parameter data corrects for interference from environmental factors on the monitoring results. The dual-branch deep neural network automatically learns high-dimensional abstract features, replacing traditional manual feature engineering, and can capture subtle differences imperceptible to the human eye. By directly mapping the fused feature vectors to purity values, error accumulation in multi-stage processing is avoided, significantly shortening the monitoring cycle. Real-time monitoring provides immediate feedback, allowing for timely adjustments to the recycling process, improving efficiency. Automated monitoring reduces manual intervention and lowers costs. This AI-based method for monitoring the purity of recycled plastic materials improves the accuracy and efficiency of monitoring the purity of recycled plastic materials, solving the problems of low efficiency and large errors in manual testing. Attached Figure Description

[0062] Figure 1 This is a flowchart of an artificial intelligence-based method for monitoring the purity of recycled plastic materials as described in this application. Detailed Implementation

[0063] like Figure 1 As shown, the method for monitoring the purity of recycled plastic materials based on artificial intelligence according to this application includes the following steps:

[0064] S1. Obtain multiple batches of historically monitored plastic recycled materials, and collect surface visual image data, spectral signal data, environmental parameter data, and corresponding plastic recycled material purity data of the plastic recycled materials. Perform data processing to obtain a plastic feature dataset. The plastic recycled material purity data includes the purity value of each plastic recycled material, which is the purity value that has been monitored in history.

[0065] S2. Construct a dual-branch deep neural network model, extract surface visual image feature vectors and spectral signal feature vectors from surface visual image data and spectral signal data respectively, and fuse them to obtain the fused feature vector of plastic recycling material;

[0066] S3. Combining the feature vector of recycled plastic materials, the dual-branch deep neural network model is trained and optimized using the plastic feature dataset to obtain the purity monitoring model of recycled plastic materials.

[0067] S4. Collect real-time surface visual image data, real-time spectral signal data, and real-time environmental parameter data of the plastic recycled material to be monitored, and input them into the plastic recycled material purity monitoring model to obtain the real-time purity value of the plastic recycled material and generate the plastic recycled material purity monitoring results.

[0068] Furthermore, in this embodiment, step S1 specifically includes:

[0069] The surface visual image data includes images of the surface of recycled plastic and plastic color information. The plastic color information is the hue, saturation, and brightness information of the corresponding HSV color space obtained from the RGB color value of each pixel in the surface image of recycled plastic.

[0070] Spectral signal data includes near-infrared spectral data of recycled plastic materials in the wavelength range of 780-2526 nm and 200-3500 cm⁻¹. -1 Raman spectral data in the wavenumber range;

[0071] Environmental parameter data includes temperature, humidity, and light intensity data of the environment in which the recycled plastic material is located;

[0072] The collected surface visual image data, spectral signal data, and environmental parameter data are transmitted to the edge computing gateway;

[0073] Specifically, for the acquisition of surface visual image data, industrial cameras, such as CCD cameras, can be used to acquire images of the surface of recycled plastic materials. Image processing algorithms can then be used to convert the RGB color values ​​of each pixel in the image of the recycled plastic material surface to the HSV color space to obtain the corresponding hue, saturation, and brightness information.

[0074] For the acquisition of spectral signal data, a near-infrared spectrometer can be used to collect near-infrared spectral data of recycled plastic materials in the wavelength range of 780-2526 nm, and a Raman spectrometer can be used to collect data in the wavelength range of 200-3500 cm⁻¹. -1 Raman spectral data in the wavenumber range;

[0075] Temperature data is collected using a temperature sensor to measure ambient temperature values; humidity data is collected using a humidity sensor to measure the relative humidity percentage; both temperature and humidity data can be collected using a temperature and humidity sensor; and light intensity data is collected using a light intensity sensor to measure the ambient lux light intensity values.

[0076] The collected surface visual image data, spectral signal data, and environmental parameter data are transmitted to the edge computing gateway via wired or wireless transmission methods.

[0077] Furthermore, in this embodiment, step S1 further includes:

[0078] The surface image of the recycled plastic material is obtained from the edge computing gateway, and then processed by median filtering for noise reduction and histogram equalization. The surface image of the recycled plastic material is converted into a grayscale image and the grayscale co-occurrence matrix is ​​calculated.

[0079] The texture pattern of the surface image of recycled plastic is extracted using local binary mode to obtain the texture features of the surface of recycled plastic.

[0080] The contour of the plastic recycled material is obtained by edge detection algorithm, and the perimeter, area, roundness and rectangularity of the contour are calculated. The shape features are extracted by Fourier descriptor.

[0081] By detecting and analyzing images of the surface of recycled plastic materials, the length and width of scratches, the length and direction of cracks, the area and diameter of pores, and the number and volume of bubbles on the surface of recycled plastic materials are obtained.

[0082] Near-infrared spectral data were filtered, smoothed, and subjected to principal component analysis for dimensionality reduction to obtain absorbance in the wavelength range of 780-2526 nm, as well as absorbance peaks at wavelengths of 1100 nm, 1450 nm, and 1900 nm.

[0083] Obtain Raman spectra from 200-3500 cm⁻¹ -1 Scattered light intensity within the wavenumber range, and polyethylene 2840cm -1 Polypropylene 800-850cm -1 Polyvinyl chloride 600-700cm -1 The intensity and position of characteristic peaks;

[0084] By combining temperature data, humidity data, and light intensity data, the processed surface visual image data and spectral signal data are calibrated. The calibrated surface visual image data, spectral signal data, environmental parameter data, and plastic recycling purity data are then correlated, labeled, and integrated to form a plastic feature dataset.

[0085] Specifically, after obtaining the surface image of the plastic recycled material from the edge computing gateway, median filtering is first performed to remove noise. Median filtering can effectively remove discrete noise points such as salt and pepper noise in the image, making the image smoother. Then, histogram equalization is performed. By adjusting the grayscale histogram of the image, the grayscale values ​​of the image are redistributed to enhance the overall contrast of the image.

[0086] Converting the processed color image to a grayscale image facilitates subsequent analysis. Calculating the grayscale co-occurrence matrix (GCM) can describe the joint grayscale distribution of two pixels with a specific spatial relationship in the image. By analyzing the eigenvalues ​​of the GCM, such as the image's energy, entropy, contrast, and correlation, the image's texture features can be obtained.

[0087] Local Binary Pattern (LBP) is used to extract the texture pattern of the surface image of recycled plastic. The basic principle of LBP is to set a circular neighborhood (e.g., radius 1, containing 8 neighboring points) centered on a certain pixel in the image. The gray value of the center pixel is used as a threshold and compared with the gray values ​​of the neighboring pixels. If the gray value of the neighboring pixels is greater than or equal to the gray value of the center pixel, it is recorded as 1; otherwise, it is recorded as 0. These comparison results are combined into a binary number in a certain order (e.g., clockwise), and then converted into a decimal number as the LBP value of the center pixel.

[0088] By calculating the LBP values ​​of all pixels in the image, the LBP texture mode of the image is obtained, thereby further obtaining the texture features of the surface of the recycled plastic material;

[0089] Edge detection algorithms (such as the Canny edge detection algorithm) are used to obtain the outline of the plastic recycling material, calculate the perimeter, area, roundness and rectangularity parameters of the outline, and extract shape features using Fourier descriptors. The Fourier descriptor transforms the coordinate information of the object outline into the frequency domain through Fourier transform and uses a series of Fourier coefficients to describe the shape of the object. These coefficients can reflect the shape characteristics of the object. Even if the object is translated, rotated and scaled, the consistency of the shape description can be maintained by properly processing the Fourier coefficients.

[0090] The surface images of recycled plastic materials are inspected and analyzed. Using image recognition and measurement algorithms, detailed features such as scratch length and width, crack length and direction, hole area and diameter, and number and volume of bubbles are determined. For example, for scratch detection, the edge of the scratch can be determined by edge detection and morphological operations, and its length and width can be measured by pixel counting. For holes, the hole area can be marked by a region growing algorithm, and then its area and diameter can be calculated.

[0091] Near-infrared spectral data can be filtered and smoothed using the Savitzky-Golay filtering method, followed by principal component analysis (PCA) dimensionality reduction. By performing eigenvalue decomposition on the covariance matrix of the spectral data, the most significant direction of change in the data can be identified, projecting the high-dimensional spectral data into a low-dimensional space. This reduces the data dimensionality and computational load while preserving most of the data information.

[0092] The absorbance in the wavelength range of 780-2526nm, as well as the absorbance peaks at wavelengths of 1100nm, 1450nm and 1900nm, are obtained. The absorbance peaks at these specific wavelengths are important indicators for identifying the type and purity of plastics.

[0093] Obtain Raman spectra from 200-3500 cm⁻¹ -1 Scattered light intensity within the wavenumber range, and polyethylene 2840cm-1 Polypropylene 800-850cm -1 Polyvinyl chloride 600-700cm -1 The intensity and position of characteristic peaks: Different plastics have unique characteristic peaks in Raman spectra. By measuring the intensity and position of these characteristic peaks, the composition of the plastic and the presence of impurities can be determined, thus providing a basis for purity analysis.

[0094] By combining temperature data, humidity data, and light intensity data, the processed surface visual image data and spectral signal data are calibrated. This is because environmental factors (such as temperature changes that may affect the physical morphology of plastics and thus affect spectral characteristics, and humidity and light intensity that may affect image acquisition quality) can interfere with the data. Therefore, the visual and spectral data need to be adjusted according to environmental parameters.

[0095] The calibrated surface visual image data, spectral signal data, environmental parameter data and plastic recycling purity data are correlated, labeled and integrated to assign a clear correspondence to each type of data for each sample, and finally form a plastic feature dataset.

[0096] An example of step S1 is as follows:

[0097] Twenty historical batches of recycled PE plastic were obtained, with 100 samples in each batch, for a total of 2,000 samples.

[0098] Using an industrial CCD camera with a resolution of 6 million pixels, surface images of 2,000 samples were captured one by one in an environment with a distance of 25 cm from the plastic recycling material and an illumination intensity of 1,300 Lux.

[0099] A near-infrared spectrometer with a wavelength range of 780-2526 nm and 6 scans was used, along with a wavenumber range of 200-3500 cm⁻¹. -1 A Raman spectrometer with a laser power of 90mW was used to acquire spectral signal data for each sample.

[0100] A temperature and humidity sensor with a measurement accuracy of ±0.4℃ and ±1.8% RH was used to record an ambient temperature of 27℃ and a humidity of 40% RH. A light intensity sensor with a measurement accuracy of ±3% FS was used to record a light intensity of 1300 Lux.

[0101] The purity data for these 20 batches of recycled plastic have been determined using conventional methods (such as chromatographic analysis). For example, the purity of the first batch was 91%, and the purity of the second batch was 89%.

[0102] After the collected data is transmitted to the edge computing gateway, the surface visual image is processed: first, a 3×3 window midpoint filter is used to remove noise, and histogram equalization is used to enhance contrast; then, the image is converted to grayscale and the gray-level co-occurrence matrix is ​​calculated.

[0103] The texture pattern was extracted using the LBP algorithm with a neighborhood radius of 1 and a neighborhood number of 8. The contour was obtained using Canny edge detection (high and low thresholds of 0.35 and 0.12). Shape parameters such as perimeter and area were calculated, and scratches with a length of 3 mm were detected in some samples.

[0104] In terms of spectral data processing, the near-infrared spectrum was smoothed by Savitzky-Golay filtering with a window size of 5, and the absorbance peak at a specific wavelength was obtained by principal component analysis for dimensionality reduction.

[0105] Extracting Raman spectra from 200-3500 cm⁻¹ -1 Scattered light intensity within the wavenumber range, and the characteristic peak of PE (2840 cm⁻¹). -1 The intensity and location of )

[0106] Visual and spectral data were calibrated by combining environmental parameters such as light intensity, temperature, and humidity. The calibrated image features, spectral features, and environmental parameters were then associated with the purity values ​​of each sample in the corresponding batch and integrated to form a plastic feature dataset containing 2,000 records.

[0107] Furthermore, in this embodiment, step S2 specifically includes:

[0108] A dual-branch deep neural network model is constructed, consisting of a surface visual image data processing branch and a spectral signal data processing branch. The surface visual image data processing branch consists of multiple convolutional layers, pooling layers, and fully connected layers, while the spectral signal data processing branch consists of multiple one-dimensional convolutional layers, global average pooling layers, and fully connected layers.

[0109] The surface visual image data in the plastic feature dataset is input into the surface visual image data processing branch, and then passes through the convolutional layer, pooling layer and fully connected layer in sequence to obtain the surface visual image feature vector. The surface visual image data is then processed by the convolutional layer for feature extraction, the pooling layer to reduce the data dimensionality, and the fully connected layer to integrate the features to obtain the surface visual image feature vector.

[0110] The spectral signal data from the plastic feature dataset is input into the spectral signal data processing branch, and then passes through a one-dimensional convolutional layer, a global average pooling layer, and a fully connected layer to obtain the spectral signal feature vector. The spectral signal data is then passed through a one-dimensional convolutional layer to extract local features, a global average pooling layer to aggregate features, and a fully connected layer to further adjust and combine the features to obtain the spectral signal feature vector.

[0111] The surface visual image feature vector and the spectral signal feature vector are fused together to form a fused feature vector of recycled plastic material.

[0112] An example of step S2 is as follows:

[0113] Construct a two-branch deep neural network model for the surface visual image data processing branch:

[0114] Convolutional layer settings: Determine the number of convolutional layers, for example, set it to 3 convolutional layers, each convolutional layer is equipped with a different number of convolutional kernels, the first convolutional layer is set to 32 3×3 convolutional kernels, stride is 1, padding is 1, and the ReLU activation function is used, its function is to extract the basic features of the image through convolution operation, such as edges, texture and other primary information;

[0115] The second convolutional layer has 64 3×3 convolutional kernels with a stride of 1 and padding of 1. It also uses the ReLU activation function to further extract more complex features.

[0116] The third convolutional layer has 128 3×3 convolutional kernels with a stride of 1, padding of 1, and ReLU activation function to make feature extraction more in-depth.

[0117] Pooling layer settings: Two pooling layers are set up, using max pooling. The first pooling layer has a window size of 2×2 and a stride of 2. Its function is to downsample the output of the convolutional layer, reduce the amount of data, and retain the main features. The second pooling layer has a window size of 2×2 and a stride of 2, which further reduces the data dimensionality.

[0118] Fully connected layer setup: Two fully connected layers are set up. The first fully connected layer has 256 neurons and uses the ReLU activation function to integrate and transform the features output by the pooling layer. The second fully connected layer has 128 neurons and prepares for the subsequent output feature vector.

[0119] For the spectral signal data processing branch:

[0120] One-dimensional convolutional layer settings: Determine the number of one-dimensional convolutional layers, for example, set it to 3 one-dimensional convolutional layers. The first one-dimensional convolutional layer is equipped with 32 convolutional kernels of length 3 and stride 1. The ReLU activation function is used to perform preliminary feature extraction on the spectral signal data.

[0121] The second one-dimensional convolutional layer has 64 convolutional kernels with a length of 3, a stride of 1, and a ReLU activation function to further extract features of the spectral signal;

[0122] The third one-dimensional convolutional layer has 128 convolutional kernels of length 3, a stride of 1, and a ReLU activation function to deeply mine spectral features;

[0123] Global average pooling layer setting: The output of the one-dimensional convolutional layer is processed by the global average pooling layer, and the average value of each feature map is used as the output, which reduces the data dimensionality and highlights the overall features;

[0124] Fully connected layer setup: Two fully connected layers are set up. The first fully connected layer has 256 neurons and uses the ReLU activation function to integrate and transform the pooled spectral features. The second fully connected layer has 128 neurons and prepares the output spectral signal feature vector.

[0125] Processing surface visual image data:

[0126] The surface visual image data in the plastic feature dataset is preprocessed, and the image size is adjusted to a uniform specification, such as 224×224 pixels, and then input into the surface visual image data processing branch.

[0127] The image is processed sequentially through the first convolutional layer, where convolutional kernels are used to perform convolution operations with the image to extract preliminary features. Then, the ReLU activation function is used to add non-linearity. Next, the image is downsampled in the first pooling layer. This process is repeated through subsequent convolutional and pooling layers to gradually extract and filter features. Finally, the image is processed through a fully connected layer to integrate and transform the features, resulting in a surface visual image feature vector of length 128.

[0128] Processing spectral signal data:

[0129] The spectral signal data in the plastic feature dataset is normalized to make its numerical range uniform, and then input into the spectral signal data processing branch.

[0130] The signal passes through the first one-dimensional convolutional layer, where the convolutional kernel performs a convolution operation with the spectral signal to extract preliminary features. Then, the ReLU activation function is used to increase nonlinearity. After the global average pooling layer reduces the dimensionality, the feature vector is then integrated and transformed through a fully connected layer to obtain a spectral signal feature vector of length 128.

[0131] Fusion feature vectors:

[0132] The obtained surface visual image feature vector and spectral signal feature vector are fused by splicing them in terms of dimensions. For example, two 128-dimensional vectors are spliced ​​into a 256-dimensional plastic recycling feature vector.

[0133] For example, the plastic feature dataset contains 2000 samples. For each sample, surface visual image data is extracted. After the size of the extracted surface visual image data is adjusted to 224×224 pixels, it is input into the surface visual image data processing branch.

[0134] Features are extracted and filtered sequentially through convolutional and pooling layers, and then passed through a fully connected layer to obtain a 128-dimensional surface visual image feature vector. At the same time, the spectral signal data of each sample is normalized and input into the spectral signal data processing branch. After passing through a one-dimensional convolutional layer, a global average pooling layer, and a fully connected layer, a 128-dimensional spectral signal feature vector is obtained. Finally, these two 128-dimensional vectors are concatenated to form a 256-dimensional fused feature vector of recycled plastic material.

[0135] Furthermore, in this embodiment, step S3 specifically includes:

[0136] The plastic feature dataset is divided into training and testing sets in an A1:A2 ratio, and the parameters of each layer of the dual-branch deep neural network model are initialized.

[0137] Set the number of training rounds Q, combine the feature vector of plastic recycling material, and use the training set to train the dual-branch deep neural network model for Q iterations.

[0138] The test set is input into the trained dual-branch deep neural network model for testing. The fused feature vector of the plastic recycled material is combined for forward propagation to obtain the predicted purity value of the plastic recycled material for all samples in the test set.

[0139] The dual-branch deep neural network model was evaluated based on the predicted purity values ​​of recycled plastic materials.

[0140] Specifically, when dividing the training set and the test set according to the ratio of A1:A2, stratified sampling can be considered. Stratified sampling can ensure that the distribution of the training set and the test set in each category is similar, and avoid model training bias caused by uneven data distribution.

[0141] For initializing the parameters of each layer in a two-branch deep neural network model, a random initialization method (such as Xavier initialization) can be selected.

[0142] Furthermore, in this embodiment, the model evaluation includes:

[0143] The purity prediction values ​​for recycled plastics include a first purity prediction value P based solely on surface visual image features. v The second purity prediction value P based solely on spectral signal characteristics s And the third purity prediction value P of the fusion feature. f Calculate the mean intermodal consistency score C. J and weighted evaluation index M C :

[0144]

[0145] Among them, C iP represents the intermodal consistency score of the i-th sample in the test set; v (i) P represents the first purity prediction value for the i-th sample in the test set; s (i) P represents the predicted second purity value for the i-th sample in the test set. f (i) y represents the predicted third purity value for the i-th sample in the test set; (i) This represents the actual purity value of the i-th sample in the test set; N represents the total number of samples in the test set; i = 1, 2, ..., N;

[0146] Based on the intermodal consistency score C i The sample prediction variance of the test set is used to dynamically shrink the confidence interval, and the confidence interval is calculated:

[0147]

[0148] Where Z represents the dynamic adjustment coefficient for the dynamically shrinking confidence interval; Z0 represents the standard confidence level quantile; and γ represents the adaptive coefficient. This represents the prediction variance calculated using Monte Carlo methods for the i-th sample in the test set; max(σ) 2 CI represents the maximum variance of all samples in the test set; i This represents the confidence interval for the i-th sample in the test set;

[0149] If M C ≤0.5, C J ≥0.7 and CI i If the proportion of samples not covering the true value is ≤20%, then the dual-branch deep neural network model is deemed to meet the requirements and will be used as the purity monitoring model for recycled plastics.

[0150] If M C ≤0.5, C J ≥0.7 and CI i If any one of the following conditions is not met: the proportion of samples not covering the true value is ≤20%, then the dual-branch deep neural network model is deemed unsuitable, and model optimization is performed until M is reached. C ≤0.5, C J ≥0.7 and CI i The requirement is met if the proportion of samples not covering the true value is ≤20%.

[0151] Furthermore, in this embodiment, model optimization includes:

[0152] Based on the obtained weighted evaluation index M C Mean of intermodal consistency score C J Confidence interval CI iThe dual-branch deep neural network model was analyzed and optimized.

[0153] When M C If the value is greater than 0.5, it is determined that feature extraction is insufficient, and a feature extraction optimization strategy is executed until M is satisfied. C ≤0.5; Feature extraction optimization strategies include increasing the number of layers or neurons in a bi-branch deep neural network to enhance the model's expressive power;

[0154] When C J If the value is less than 0.7, the feature fusion mechanism is deemed to have a defect, and an optimization strategy for the feature fusion mechanism is executed until C is satisfied. J ≥0.7; The feature fusion mechanism optimization strategy includes adding a cross-modal contrastive learning module in the feature extraction stage, which enhances modal consistency by minimizing the distance between the two modal features in the embedding space;

[0155] When CI i If the proportion of samples not covering the true value is greater than 20%, then the confidence interval is determined to be unable to effectively contain the true value, and a confidence interval optimization strategy is executed until the CI is met. i The proportion of samples not covering the true value is ≤20%; the confidence interval optimization strategy includes fine-tuning the calculation formula of the dynamic adjustment coefficient z based on the deviation between the actual prediction results and the true value, and optimizing the confidence interval generation mechanism.

[0156] An example of step S3 is as follows:

[0157] The dataset contains 2000 samples, all of which are recycled PE (polyethylene) plastic.

[0158] The dataset is divided into a training set and a test set according to the ratio of A1:A2 = 8:2. The training set contains 2000 × 0.8 = 1600 samples, and the test set contains 2000 × 0.2 = 400 samples.

[0159] During the segmentation process, a stratified sampling method was adopted, which was used to segment the plastic recycled materials according to different purity ranges, production batches and other factors to ensure that the sample distribution of the training set and the test set was similar in all dimensions.

[0160] The Xavier initialization method is used to initialize the parameters of each layer of the dual-branch deep neural network model. The surface vision image data processing branch is constructed as 4 convolutional layers, 2 pooling layers, and 2 fully connected layers. The first convolutional layer is set with 32 3×3 convolutional kernels, stride 1, and padding 1.

[0161] The spectral signal data processing branch is constructed as 3 one-dimensional convolutional layers, 1 global average pooling layer, and 2 fully connected layers. The first one-dimensional convolutional layer is set with 32 convolutional kernels of length 3 and stride 1.

[0162] The training rounds are set to Q=60, the Adam optimization algorithm is selected, the initial learning rate is set to 0.001, and the feature vector of plastic recycling is fused. The dual-branch deep neural network model is trained using the training set. The loss function is calculated in each training round. If the loss function decreases by less than 0.0003 for 4 consecutive rounds, the training is terminated early.

[0163] The data from 400 samples in the test set were preprocessed. The surface visual image data was normalized and adjusted to 224×224 pixels, and the spectral signal data was standardized. The processed data was then input into a trained dual-branch deep neural network model. The feature vector fused with the plastic recycled material was then propagated forward. For example, sample P-001 was processed by the surface visual image data processing branch and the spectral signal data processing branch, respectively. After fusion, the data was output through a fully connected layer, resulting in a purity prediction value of 87%. This process was repeated for all 400 samples to obtain the predicted value.

[0164] Calculate the intermodal consistency score C i Taking sample P-002 as an example, the first purity prediction value P is based solely on surface visual image features. v (002) =85%, the second purity prediction value P based solely on spectral signal characteristics s (002) =86%, according to the formula We can get C 002 =0.99, calculate C for 400 samples. i The average of the intermodal consistency scores, C, is then calculated. J =0.8;

[0165] Calculate the weighted evaluation index M C Given the third purity prediction value P of sample P-002 based on fusion features. f (002) =88%, actual purity value y (002) =89%, substitute into the formula (where N = 400), after calculation, M C =0.48;

[0166] Calculate the confidence interval CI i The sample prediction variance was calculated using Monte Carlo simulation. For example, sample P-002 The standard confidence level quantile Z0 = 1.96, the adaptive coefficient γ = 0.5, and the maximum variance of all samples in the test set max(σ) 2) = 0.006, substitute into the formula to calculate the dynamic adjustment coefficient Z, and obtain the confidence interval CI2 = [86%, 90%] for sample P-002. The statistically found that the proportion of samples whose CI2 does not cover the true value is 16%.

[0167] Because M C =0.48≤0.5, C J =0.8≥0.7, CI2 does not cover the true value of 16%≤20% of the samples, so the dual-branch deep neural network model is deemed to meet the requirements and can be used as a purity monitoring model for plastic recycled materials.

[0168] Furthermore, in this embodiment, step S4 specifically includes:

[0169] Real-time surface visual image data, real-time spectral signal data, and real-time environmental parameter data of the plastic recycled material to be monitored are collected, processed, and then input into the plastic recycled material purity monitoring model to obtain the preliminary purity value P of the plastic recycled material. pre The data collection and processing methods and steps for the recycled plastic materials to be monitored are consistent with those in S1.

[0170] Based on a pre-established calibration sample library, a similarity matching algorithm is used to retrieve multiple historical samples that are most similar to the characteristics of the current recycled plastic material, and the mean calibration purity value P is calculated. J And the initial purity value P pre The real-time purity value P of the recycled plastic material was obtained through calibration. S :

[0171] P S =P pre +α*(P J -P pre )

[0172] Wherein, α represents the adaptive weight, which is dynamically adjusted according to the sample matching degree, and its value ranges from 0 to 1; the adaptive weight α is selected according to the matching degree, and a larger α is selected when the matching degree is high, so as to make the mean calibration purity P J For the initial purity value P pre The calibration effect is stronger. When the matching degree is low, a smaller α is selected, which can reduce the calibration amplitude and avoid over-calibration.

[0173] Based on the real-time purity value P S The purity of recycled plastic materials is analyzed, and the purity monitoring results of recycled plastic materials are generated.

[0174] Furthermore, in this embodiment, the purity analysis includes:

[0175] If P SIf the purity is ≥95%, the plastic recycling material is determined to be of grade one purity, and the plastic recycling material will continue to undergo the original processing flow. When the purity grade is grade one, it means that the impurity content of the plastic recycling material is extremely low and does not affect its physical and chemical properties. Therefore, no additional processing is required, and the subsequent operations of the original processing flow can be carried out.

[0176] If 90% ≤ P S If the purity is less than 95%, the plastic recycling material is determined to be of grade two, and an alarm message is issued, requiring a second re-inspection. A grade two purity indicates that the plastic recycling material contains a small amount of impurities, which may have a certain impact on the quality of subsequent products. Therefore, a second re-inspection is required to more accurately detect the type and content of impurities and to further adjust the processing technology of the plastic recycling material based on the re-inspection results.

[0177] If 80% ≤ P S If the purity is less than 90%, the plastic recycling material is determined to be of grade three, and an alarm message is issued. The plastic recycling material is then sorted to the downgrade processing area. When the purity grade is grade three, it means that the plastic recycling material has a relatively high impurity content, and its quality can hardly meet the production requirements of conventional high-quality products. Therefore, it needs to be sorted to the downgrade processing area in order to use the plastic recycling material in the production of products with lower purity requirements, so as to achieve the rational use of resources.

[0178] If P S If the purity is less than 80%, the plastic recycling material is determined to be of grade four, and an alarm message is issued. The plastic recycling material will then undergo enhanced impurity removal treatment. When the purity grade is grade four, it means that the plastic recycling material contains a large number of impurities and cannot be directly used for the production of plastic products. Therefore, enhanced impurity removal treatment can be carried out to remove impurities through enhanced impurity removal technology (such as pyrolysis technology) to improve the purity of the plastic recycling material.

[0179] Furthermore, in this embodiment, the purity monitoring results of the recycled plastic material include:

[0180] The real-time surface visual image data, real-time spectral signal data, real-time environmental parameter data and real-time purity value of the same batch of recycled plastic materials are integrated and processed, and the purity monitoring results of recycled plastic materials containing real-time purity value, purity level and purity fluctuation trend are generated through visualization technology.

[0181] An example of step S4 is as follows:

[0182] Real-time monitoring of the PE plastic recycling material was carried out using an industrial CCD camera with a resolution of 6 million pixels. The surface visual image data of the plastic recycling material was collected in real time at a distance of 25cm from the plastic recycling material and an illumination intensity of 1300Lux.

[0183] A near-infrared spectrometer with a wavelength range of 780-2526 nm and 6 scans was used, along with a wavenumber range of 200-3500 cm⁻¹. -1 A Raman spectrometer with a laser power of 90mW was used to acquire real-time spectral signal data.

[0184] Using a temperature and humidity sensor with a measurement accuracy of ±0.4℃ and ±1.8% RH, the real-time ambient temperature of 27℃ and humidity of 40% RH were recorded. A set of data was collected every 2 minutes, and each set of data contained relevant data of several plastic recycled material samples that passed through the production line during that time period.

[0185] For the acquired real-time surface visual image data, a 3×3 window midpoint filter is first used to remove noise, and histogram equalization is used to enhance contrast.

[0186] After converting the image to grayscale, the grayscale co-occurrence matrix is ​​calculated, and the texture pattern is extracted using the LBP algorithm with a neighborhood radius of 1 and 8 neighborhood points.

[0187] The contour was obtained using Canny edge detection (with high and low thresholds of 0.35 and 0.12, respectively), and the perimeter and area were calculated.

[0188] For real-time spectral signal data, the near-infrared spectrum is smoothed by Savitzky-Golay filtering with a window size of 5, and the absorbance peak at a specific wavelength is obtained by dimensionality reduction through principal component analysis.

[0189] Extracting Raman spectra from 200-3500 cm⁻¹ -1 Scattered light intensity within the wavenumber range, and the characteristic peak of PE (2840 cm⁻¹). -1 The intensity and location of )

[0190] After calibrating the visual and spectral data using real-time environmental parameters (including a light intensity of 1300 Lux), the processed data is input into the trained and compliant plastic recycling purity monitoring model in step S3 to obtain the preliminary purity value P of the plastic recycling. pre For example, after processing the collected data and inputting it into the model, a preliminary purity value P is obtained. pre =86%;

[0191] A calibration sample library containing 500 historical samples was pre-established. These samples cover PE plastic recycling materials with different purity ranges and production batches. A cosine similarity matching algorithm was used to retrieve the five historical samples most similar to the current plastic recycling material. For example, the purity values ​​of these five historical samples were 86%, 89%, 90%, 87%, and 88%, respectively. The mean calibration purity was calculated.

[0192] The adaptive weight α is dynamically adjusted based on the sample matching degree. For example, if the matching degree is high, α is set to 0.6, according to the formula P. S =88% + 0.6 * (88% - 86%) = 89.2%;

[0193] Because 80% ≤ P S If the purity of the recycled plastic is 89.2% < 90%, then the purity level of the recycled plastic is determined to be level three, an alarm message is issued, and the recycled plastic is sorted to the downgrade processing area.

[0194] The real-time surface visual image data, real-time spectral signal data, real-time environmental parameter data, and real-time purity value of each piece of plastic recycled material in the batch to be monitored are integrated and processed.

[0195] The current batch of 150 pieces of recycled plastic is to be monitored. After monitoring all 150 pieces, the number of pieces at each purity level will be counted: 50 pieces at level 1 purity, 30 pieces at level 2 purity, 50 pieces at level 3 purity, and 20 pieces at level 4 purity. The monitoring results of these 150 pieces of recycled plastic will be visualized using data visualization technology. A line graph will show the purity fluctuation trend of the 150 monitored materials, and a bar chart will show the quantity distribution of each purity level. At the same time, the real-time purity value, purity level, overall monitoring progress and abnormal status of each piece of recycled plastic will be marked.

[0196] In the description of this application, it should be understood that the orientation or positional relationship indicated by directional terms such as "front, back, up, down, left, right", "horizontal, vertical, horizontal" and "top, bottom" is usually based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing this application and simplifying the description. Unless otherwise stated, these directional terms do not indicate or imply that the device or element referred to must have a specific orientation or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on the scope of protection of this application.

[0197] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this application.

Claims

1. A method for monitoring the purity of recycled plastic materials based on artificial intelligence, characterized in that, Includes the following steps: S1. Obtain historical batches of recycled plastic with monitored purity, and collect surface visual image data, spectral signal data, environmental parameter data, and corresponding plastic purity data of the recycled plastic. Perform data processing to obtain a plastic feature dataset. The environmental parameter data includes temperature data, humidity data, and light intensity data of the environment in which the recycled plastic is located. Combine the temperature data, humidity data, and light intensity data to calibrate the processed surface visual image data and spectral signal data. Associate, label, and integrate the calibrated surface visual image data, spectral signal data, environmental parameter data, and plastic purity data with the recycled plastic data to form a plastic feature dataset. S2. Construct a dual-branch deep neural network model, extract surface visual image feature vectors and spectral signal feature vectors from the surface visual image data and the spectral signal data respectively, and fuse them to obtain the fused feature vector of the plastic recycled material; S3. Combining the fused feature vector of the plastic recycled material, the dual-branch deep neural network model is trained and optimized using the plastic feature dataset to obtain the plastic recycled material purity monitoring model. S4. Collect real-time surface visual image data, real-time spectral signal data, and real-time environmental parameter data of the plastic recycled material to be monitored, process the data, and input the processed data into the plastic recycled material purity monitoring model to obtain the real-time purity value of the plastic recycled material and generate the plastic recycled material purity monitoring result.

2. The method for monitoring the purity of recycled plastic materials based on artificial intelligence according to claim 1, characterized in that, Step S1 specifically includes: The surface visual image data includes a surface image of recycled plastic and plastic color information. The plastic color information is the hue, saturation, and brightness information of the corresponding HSV color space obtained based on the RGB color value of each pixel in the surface image of recycled plastic. The spectral signal data includes near-infrared spectral data of recycled plastic materials in the wavelength range of 780-2526 nm and 200-3500 cm⁻¹. -1 Raman spectral data in the wavenumber range; The environmental parameter data includes temperature, humidity, and light intensity data of the environment in which the recycled plastic material is located; The collected surface visual image data, spectral signal data, and environmental parameter data are transmitted to the edge computing gateway.

3. The method for monitoring the purity of recycled plastic materials based on artificial intelligence according to claim 2, characterized in that, Step S1 further includes: The surface image of the recycled plastic material is obtained from the edge computing gateway, and then subjected to median filtering for noise reduction and histogram equalization. The surface image of the recycled plastic material is converted into a grayscale image and the grayscale co-occurrence matrix is ​​calculated. The texture pattern of the surface image of the recycled plastic material is extracted using a local binary mode to obtain the texture features of the surface of the recycled plastic material; The contour of the plastic recycled material is obtained by edge detection algorithm, and the perimeter, area, roundness and rectangularity of the contour are calculated. The shape features are extracted by Fourier descriptor. The surface image of the recycled plastic is detected and analyzed to obtain the scratch length and width, crack length and direction, hole area and diameter, and number and volume of bubbles on the surface of the recycled plastic. The near-infrared spectral data were filtered, smoothed, and subjected to principal component analysis for dimensionality reduction to obtain absorbance in the wavelength range of 780-2526 nm, as well as absorbance peaks at wavelengths of 1100 nm, 1450 nm, and 1900 nm. Obtain Raman spectra from 200 to 3500 cm⁻¹ -1 Scattered light intensity within the wavenumber range, and polyethylene 2840cm -1 Polypropylene 800-850cm -1 Polyvinyl chloride 600-700cm -1 The intensity and position of characteristic peaks; By combining the temperature data, humidity data, and light intensity data, the processed surface visual image data and spectral signal data are calibrated. The calibrated surface visual image data, spectral signal data, environmental parameter data, and plastic recycling purity data are then correlated, labeled, and integrated to form a plastic feature dataset.

4. The method for monitoring the purity of recycled plastic materials based on artificial intelligence according to claim 3, characterized in that, Step S2 specifically includes: A dual-branch deep neural network model is constructed, comprising a surface visual image data processing branch and a spectral signal data processing branch. The surface visual image data processing branch consists of multiple convolutional layers, pooling layers, and fully connected layers, while the spectral signal data processing branch consists of multiple one-dimensional convolutional layers, global average pooling layers, and fully connected layers. The surface visual image data in the plastic feature dataset is input into the surface visual image data processing branch, and then passed through the convolutional layer, pooling layer and fully connected layer in sequence to obtain the surface visual image feature vector; The spectral signal data in the plastic feature dataset is input into the spectral signal data processing branch, and then passed through a one-dimensional convolutional layer, a global average pooling layer, and a fully connected layer in sequence to obtain the spectral signal feature vector. The surface visual image feature vector and the spectral signal feature vector are fused together to form a plastic recycled material fused feature vector.

5. The method for monitoring the purity of recycled plastic materials based on artificial intelligence according to claim 4, characterized in that, Step S3 specifically includes: The plastic feature dataset is divided into training and testing sets according to the ratio A1:A2, and the parameters of each layer of the dual-branch deep neural network model are initialized. Set the training rounds Q, combine the fused feature vector of the plastic recycled material, and use the training set to train the dual-branch deep neural network model for Q iterations; The test set is input into the trained dual-branch deep neural network model for testing. The fused feature vector of the plastic recycled material is combined with forward propagation to obtain the predicted purity value of plastic recycled material for all samples in the test set. The dual-branch deep neural network model is evaluated based on the predicted purity value of the recycled plastic material.

6. The method for monitoring the purity of recycled plastic materials based on artificial intelligence according to claim 5, characterized in that, The model evaluation includes: The purity prediction value of the recycled plastic includes a first purity prediction value based solely on surface visual image features. Second purity prediction value based solely on spectral signal characteristics and the third purity prediction value of fusion characteristics. Calculate the mean intermodal consistency score. and weighted evaluation indicators : ， ， ， in, Indicates the first test set Intermodal consistency score of each sample; Indicates the first test set The first purity prediction value for each sample; Indicates the first test set The second purity prediction value for each sample; Indicates the first test set The third purity prediction value for each sample; Indicates the first test set The actual purity value of each sample; This indicates the total number of samples in the test set; =1, 2…N; Based on the intermodal consistency score The sample prediction variance of the test set is used to dynamically shrink the confidence interval, and the confidence interval is calculated: ， ， in, This represents the dynamic adjustment coefficient for dynamically shrinking the signal interval; Indicates the standard confidence level quantile; Indicates the adaptive coefficient; Indicates the first test set The predicted variance of each sample was calculated using Monte Carlo. This represents the maximum variance of all samples in the test set; Indicates the first test set Confidence intervals for each sample; like ≤0.5、 ≥0.7 and If the proportion of samples not covering the true value is ≤20%, then the dual-branch deep neural network model is deemed to meet the requirements, and the dual-branch deep neural network model is used as the purity monitoring model for plastic recycled materials. like ≤0.5、 ≥0.7 and If any one of the following conditions is not met: the proportion of samples not covering the true value is ≤20%, then the dual-branch deep neural network model is deemed unsuitable, and model optimization is performed until... ≤0.5、 ≥0.7 and The requirement that the proportion of samples not covering the true value is ≤20% is met.

7. The method for monitoring the purity of recycled plastic materials based on artificial intelligence according to claim 6, characterized in that, The model optimization includes: Based on the obtained weighted evaluation indicators The mean of the intermodal consistency score and confidence interval The dual-branch deep neural network model is analyzed and optimized. when If the value is greater than 0.5, it is determined that feature extraction is insufficient, and a feature extraction optimization strategy is executed until the requirement is met. ≤0.5; when If the value is less than 0.7, the feature fusion mechanism is deemed to have a defect, and an optimization strategy for the feature fusion mechanism is executed until the condition is met. ≥0.7; when If the proportion of samples not covering the true value is greater than 20%, then the confidence interval is determined to be unable to effectively contain the true value, and a confidence interval optimization strategy is executed until the condition is met. The proportion of samples that do not cover the true value is ≤20%.

8. The method for monitoring the purity of recycled plastic materials based on artificial intelligence according to claim 7, characterized in that, Step S4 specifically includes: Real-time surface visual image data, real-time spectral signal data, and real-time environmental parameter data of the plastic recycled material to be monitored are collected, processed, and then input into the plastic recycled material purity monitoring model to obtain a preliminary purity value of the plastic recycled material. ; Based on a pre-established calibration sample library, a similarity matching algorithm is used to retrieve multiple historical samples that are most similar to the characteristics of the current recycled plastic material, and the average calibration purity is calculated. and the preliminary purity value The real-time purity value of the recycled plastic material was obtained through calibration. : ， in, This represents adaptive weights, which are dynamically adjusted based on the sample matching degree, and the value range is 0 to 1. According to the real-time purity value The purity of recycled plastic materials is analyzed, and the purity monitoring results of recycled plastic materials are generated.

9. The method for monitoring the purity of recycled plastic materials based on artificial intelligence according to claim 8, characterized in that, The purity analysis includes: like If the purity is ≥95%, the plastic recycling material is determined to be of Grade 1 purity, and the plastic recycling material will continue to undergo the original processing procedure. If 90%≤ If the purity is less than 95%, the plastic recycling material is determined to be of grade two, an alarm is issued, and the plastic recycling material is subjected to a second re-inspection. If 80%≤ If the purity is less than 90%, the plastic recycling material is determined to be of grade three, an alarm is issued, and the plastic recycling material is sorted to the downgrade processing area. like If the purity is less than 80%, the plastic recycling material is determined to be of grade four, and an alarm message is issued, requiring the plastic recycling material to undergo enhanced impurity removal treatment.

10. The method for monitoring the purity of recycled plastic materials based on artificial intelligence according to claim 9, characterized in that, The purity monitoring results of the recycled plastic materials include: The real-time surface visual image data, real-time spectral signal data, real-time environmental parameter data, and real-time purity value of the same batch of recycled plastic materials are integrated and processed, and a plastic recycled material purity monitoring result containing real-time purity value, purity level, and purity fluctuation trend is generated through visualization technology.

Images