An industrial solid waste intelligent classification method based on big data

Abstract

The application provides an industrial solid waste intelligent classification method based on big data, comprising: obtaining fly ash porosity and steel slag oxide layer data, analyzing surface micro characteristics, and extracting light scattering intensity and infrared radiation characteristics; determining the correlation representation of environmental parameters and fly ash and steel slag surface characteristics according to the nonlinear interaction relationship of humidity, temperature and surface characteristics; extracting classification feature semantics from the correlation representation of environmental parameters and fly ash and steel slag surface characteristics, and constructing a fly ash and steel slag semantic enhanced classification feature set; according to the nonlinear response characteristics of fly ash and steel slag extracted from the classification feature set, optimizing the recognition accuracy of the classification features, obtaining the optimized classification feature set, classifying the solid waste type, and generating a preliminary classification model; obtaining classification error samples from the preliminary classification model, iteratively optimizing the preliminary classification model, obtaining a final classification model, and outputting a solid waste classification result adaptive to environmental changes.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an intelligent classification method for industrial solid waste based on big data. Background Technology

[0002] Industrial solid waste treatment is a key area for promoting green manufacturing and the circular economy, its importance lying in reducing environmental pollution, improving resource utilization efficiency, and achieving sustainable development goals. With deepening industrialization, solid waste is diverse and complex in composition; improper treatment can lead to soil and water pollution and even ecological crises. Therefore, efficient and intelligent classification technologies are urgently needed to support precise resource utilization. Currently, solid waste classification methods mostly rely on single physical or chemical detection methods, such as classification based on particle size distribution or chemical composition analysis. However, these methods often exhibit instability when facing complex environmental changes, especially in dynamic environmental interactions, making it difficult to capture the nonlinear changes in solid waste surface characteristics, leading to decreased classification accuracy. For example, existing technologies struggle to accurately distinguish solid waste types whose surface characteristics change abruptly due to variations in environmental humidity, limiting the adaptability and reliability of classification systems. The core challenge in this field lies in effectively capturing the dynamic correlation between the microscopic properties of solid waste surfaces and environmental factors. The microscopic properties of solid waste surfaces, such as porosity or oxide layer state, change nonlinearly due to factors such as environmental humidity and temperature. For example, the surface porosity of fly ash can cause significant changes in light scattering intensity under humidity fluctuations, which directly affects the accuracy of optical detection. These abrupt changes in light scattering intensity further lead to instability in classification feature extraction, as existing methods struggle to establish dynamic correlations between these nonlinear responses and environmental parameters. For instance, in real-world industrial scenarios, when the humidity in a processing workshop changes rapidly from low to high, the optical signal of fly ash may exhibit abrupt changes, and existing classification systems cannot adjust their feature extraction strategies in real time, resulting in increased classification error rates. Therefore, constructing a knowledge system that can dynamically correlate environmental parameters with the surface characteristics of solid waste, and optimizing the feature extraction process to improve the classification system's ability to recognize and maintain stability against nonlinear responses, becomes a crucial issue. Summary of the Invention

[0003] This invention provides a smart classification method for industrial solid waste based on big data, mainly including:

[0004] Data on fly ash porosity and steel slag oxide layer were obtained, surface microstructures were analyzed, and light scattering intensity and infrared radiation characteristics were extracted.

[0005] Infrared spectral analysis is performed on light scattering intensity and infrared radiation characteristics to obtain spectral data. Characteristic peaks of the target band are extracted from the preprocessed spectral data. Peak shifts and intensity changes caused by changes in environmental parameters are calculated to obtain the characteristic shift under environmental parameter changes.

[0006] For the feature offset, a dynamic association knowledge graph of surface properties of fly ash and steel slag is constructed, a multi-node association model is trained, the dynamic change trend of feature offset captured by time series analysis is input into the multi-node association model, and the nonlinear interaction relationship between humidity, temperature and surface properties is output.

[0007] The correlation between environmental parameters and the surface properties of fly ash and steel slag is determined based on the nonlinear interaction between humidity, temperature and surface properties.

[0008] Classification feature semantics are extracted from the correlation representation between environmental parameters and surface properties of fly ash and steel slag, and a semantically enhanced classification feature set for fly ash and steel slag is constructed.

[0009] Based on the nonlinear response characteristics of fly ash and steel slag extracted from the classification feature set, the recognition accuracy of the classification features is optimized to obtain an optimized classification feature set, which is used to classify solid waste types and generate a preliminary classification model.

[0010] The model obtains misclassified samples from the preliminary classification model, iteratively optimizes the preliminary classification model, obtains the final classification model, and outputs solid waste classification results that adapt to environmental changes.

[0011] Furthermore, data on fly ash porosity and steel slag oxide layer were obtained, surface microstructures were analyzed, and light scattering intensity and infrared radiation characteristics were extracted, including:

[0012] Two-dimensional cross-sectional images of fly ash particles are acquired to identify pore regions and solid regions, and the porosity is obtained by calculating the ratio of pore area to total area. Energy dispersive spectroscopy (EDS) analysis is performed on the surface of steel slag samples to determine the oxide layer thickness based on the oxygen content distribution. The surfaces of fly ash and steel slag samples are irradiated with a laser, and scattered light signals are collected using a photodetector array to record the intensity of scattered light at each angle. The infrared radiation intensity value is calculated based on the correspondence between the absorption peak intensity and the oxide layer thickness. The maximum intensity value within a preset scattering angle is extracted as the light scattering intensity feature, and the average radiation intensity value within a preset wavelength range in the infrared absorption spectrum is extracted as the infrared radiation feature.

[0013] Furthermore, the infrared spectral analysis of light scattering intensity and infrared radiation characteristics is performed to obtain spectral data. Characteristic peaks of the target band are extracted from the preprocessed spectral data. Peak shifts and intensity changes caused by environmental parameter variations are calculated to obtain the characteristic shift under environmental parameter variations, including:

[0014] The light scattering intensity and infrared radiation characteristics are frequency domain transformed to obtain spectral data; the spectral data are denoised using a moving average filter, and baseline correction is performed using a polynomial fitting method to obtain a baseline-corrected spectral curve; from the baseline-corrected spectral curve, characteristic peaks in the target band are identified using the second derivative method, and the wavelength position and peak intensity of the characteristic peaks are recorded.

[0015] Furthermore, for the feature offsets, a dynamic association knowledge graph of fly ash and steel slag surface characteristics is constructed, a multi-node association model is trained, and the dynamic change trend of feature offsets captured by time series analysis is input into the multi-node association model. The output is the nonlinear interaction relationship between humidity, temperature and surface characteristics, including:

[0016] Nodes and edges of a knowledge graph are constructed, including feature offset nodes, humidity nodes, and temperature nodes, as well as fly ash surface characteristic nodes and steel slag surface characteristic nodes extracted from historical data. The weight values ​​of the edges are determined using the Pearson correlation coefficient to form the knowledge graph. The node features of the knowledge graph are aggregated and updated using a graph convolutional network to obtain node embeddings. The time series data is segmented using a sliding window method, and the mean and rate of change of the feature offsets are calculated. The node embeddings are input into a long short-term memory network, and the feature offset prediction sequence is output. The node embeddings and prediction sequences are concatenated using a multilayer perceptron to output the nonlinear interaction relationship.

[0017] Furthermore, the determination of the correlation between environmental parameters and the surface properties of fly ash and steel slag based on the nonlinear interaction between humidity, temperature, and surface properties includes:

[0018] The main effect values ​​of humidity, temperature, and humidity-temperature interaction are extracted from the nonlinear interaction relationship. The contribution ratio of each effect to the surface property change is calculated through normalization to obtain the environmental parameter coupling strength index. A three-dimensional tensor is constructed, and the tensor is decomposed to obtain feature vectors, determining the influence pattern of environmental condition combinations on surface properties. The probability density function of the influence pattern is calculated using the kernel density estimation method, dividing the environmental parameter space into different response zones and generating an environmental parameter partition mapping relationship. The representative value and frequency weight of each partition in the partition mapping relationship are calculated to obtain a comprehensive response index, and the associated representation is constructed.

[0019] Furthermore, the step of extracting classification feature semantics from the correlation representation between environmental parameters and the surface properties of fly ash and steel slag, and constructing a semantically enhanced classification feature set for fly ash and steel slag, includes:

[0020] The element values ​​of the association representation are parsed and converted into descriptive feature words to generate initial classification feature semantics; the parameter combination of the association representation is mapped into a semantic vector, and the cosine similarity between nodes is calculated to obtain a semantic similarity matrix; the nodes in the semantic similarity matrix are divided by a threshold-based semantic segmentation method to generate semantic categories and labels; the node categories containing fly ash and steel slag features are identified to form a feature subset, high-frequency feature words and basic feature words are added, and the classification feature set is output.

[0021] Furthermore, based on the nonlinear response characteristics of fly ash and steel slag extracted from the classification feature set, the accuracy of classification feature identification is optimized to obtain an optimized classification feature set. This set is then used to classify solid waste types and generate a preliminary classification model, including:

[0022] Extract the nonlinear response characteristics from the classification feature set, calculate the feature discrimination, remove redundant features, and obtain a simplified feature set; adjust the feature weights of the simplified feature set using the information gain method to generate a weighted feature vector, and construct an optimized classification feature set; construct a decision tree using the random forest algorithm, calculate the classification consistency ratio, determine stability, and output a preliminary classification model containing classification rules.

[0023] Furthermore, the step of obtaining misclassified samples from the preliminary classification model, iteratively optimizing the preliminary classification model to obtain the final classification model, and outputting solid waste classification results adapted to environmental changes includes:

[0024] Extract the erroneous samples and their environmental parameter data from the preliminary classification model to generate an erroneous sample set; calculate the residuals using the gradient boosting decision tree algorithm, iteratively optimize the preliminary classification model, and obtain the final classification model; input real-time environmental parameters, calculate the classification confidence index, and mark samples to be updated; adjust the node weights of the knowledge graph using the online gradient descent method, update the association strength values ​​and edge connections, and output the solid waste classification results.

[0025] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0026] This invention discloses an intelligent classification method for industrial solid waste based on big data. Addressing the problem of unstable classification in business scenarios where the microscopic properties of solid waste surfaces are affected by changes in humidity and temperature, the method accurately captures feature peak shifts and intensity changes through light scattering intensity and infrared radiation feature extraction, Fourier transform spectral analysis, and noise reduction preprocessing. It constructs a dynamic association knowledge graph of fly ash and steel slag surface properties, and combines graph neural networks to model nonlinear interaction relationships, revealing the correlation between environmental parameters and surface properties. Furthermore, it optimizes classification features through semantic segmentation and random forest algorithms, iteratively optimizes the classification model using gradient boosting decision trees, and introduces an online learning mechanism to dynamically update the knowledge graph, achieving accurate classification that adapts to environmental changes. This invention significantly improves the stability and accuracy of solid waste classification through a multi-node association model and dynamic update mechanism, providing efficient technical support for solid waste treatment in complex environments. Attached Figure Description

[0027] Figure 1 This is a flowchart of an intelligent classification method for industrial solid waste based on big data according to the present invention. Detailed Implementation

[0028] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] like Figure 1 This embodiment of an intelligent classification method for industrial solid waste based on big data may specifically include:

[0030] S101. Obtain data on fly ash porosity and steel slag oxide layer, analyze surface microstructure, and extract light scattering intensity and infrared radiation characteristics.

[0031] Scanning electron microscopy (SEM) was used to image fly ash samples, obtaining two-dimensional cross-sectional images of fly ash particles. Image binarization was used to identify pore regions and solid regions, and the porosity was calculated as the ratio of pore area to total area. Simultaneously, energy dispersive spectroscopy (EDS) analysis was performed on the surface of steel slag samples to determine the oxide layer thickness based on oxygen content distribution. For the obtained porosity and oxide layer thickness, a laser was used to irradiate the surface of a mixed fly ash and steel slag sample. A photodetector array was used to collect scattered light signals at different angles. The intensity of scattered light at each angle was recorded based on the correlation between scattered light intensity and scattering angle. If the porosity was greater than a preset threshold, the scattered light intensity in the corresponding area exhibited a multi-peak distribution. Based on the scattered light intensity values ​​at each angle and the multi-peak distribution characteristics, a Fourier transform infrared (FTIR) spectrometer was used to scan the samples, obtaining infrared absorption spectra in the wavelength range of 2.5-25 micrometers. The positions of characteristic absorption peaks were determined by comparing with a standard spectral library. Based on the correlation between absorption peak intensity and oxide layer thickness, the infrared radiation intensity values ​​for each band were calculated. Based on the obtained scattered light intensity values ​​at various angles, multi-peak distribution characteristics, and infrared radiation intensity values ​​in various bands, the maximum intensity value in the scattering angle range of 30-60 degrees was extracted as the light scattering intensity feature through data normalization processing, and the average radiation intensity value in the wavelength range of 8-12 micrometers was extracted as the infrared radiation feature, thus obtaining the light scattering intensity and infrared radiation characteristics corresponding to the porosity of fly ash and the oxide layer of steel slag.

[0032] Specifically, scanning electron microscopy (SEM) plays a crucial role in the microstructure analysis of fly ash.

[0033] It should be noted that the pore structure of fly ash particles directly affects their physicochemical properties. Two-dimensional cross-sectional images obtained through scanning electron microscopy can clearly show the pore distribution within the particles. Image binarization is the process of converting a grayscale image into a black-and-white binary image, where pore areas are represented in black and solid areas in white. This processing method can accurately distinguish the boundary between pores and solid areas, thereby calculating the proportion of pore area to the total area, i.e., porosity.

[0034] In one possible implementation, energy dispersive spectroscopy (EDS) is used to determine the thickness of the oxide layer on the steel slag surface. During the high-temperature smelting and cooling process, an oxide layer of varying thickness forms on the surface of the steel slag, and the presence of this oxide layer significantly affects the reactivity of the steel slag. EDS detects characteristic X-rays of elements on the sample surface, enabling the determination of the oxygen distribution curve along the depth direction. The depth at which the oxygen content gradually decreases from the surface to the interior and tends to stabilize is defined as the oxide layer thickness.

[0035] Specifically, laser scattering analysis is an important tool for studying the microscopic properties of material surfaces. When a laser beam illuminates the surface of a mixture of fly ash and steel slag, the presence of a porous structure causes complex scattering of the light. A photodetector array positioned at different angles around the sample can simultaneously collect scattered light signals from multiple directions. The relationship between the intensity of the scattered light and the scattering angle reflects the surface roughness and porosity distribution characteristics of the material. When the porosity is high, the light undergoes multiple scatterings at the pore edges, resulting in multiple peaks in the scattered light intensity within a specific angular range, forming a multi-peak distribution characteristic.

[0036] Preferably, Fourier transform infrared spectroscopy (FTIR) technology can deeply analyze the molecular structure and chemical composition of materials. The infrared light emitted by the infrared spectrometer scans the sample within a wavelength range of 2.5-25 micrometers. Different chemical bonds will absorb at specific wavelengths, forming characteristic absorption peaks. By comparing with a standard spectral library, characteristic peaks of aluminosilicate structures in fly ash and oxides on the surface of steel slag can be identified. The intensity of the absorption peaks corresponds to the thickness of the oxide layer; the thicker the oxide layer, the stronger the absorption peak of the corresponding oxide. Based on this, the infrared radiation intensity values ​​of each band can be calculated. Data normalization is the process of converting data of different dimensions to a unified standard, which allows the scattered light intensity values ​​and infrared radiation intensity values ​​to be compared and analyzed on the same scale. The maximum intensity value in the scattering angle range of 30-60 degrees best reflects the surface scattering characteristics of the material because this angle range avoids the interference of specular reflection and small-angle scattering. The wavelength range of 8-12 micrometers corresponds to the atmospheric window region, where infrared radiation is least affected by atmospheric absorption, and the average radiation intensity value best reflects the thermal radiation characteristics of the material.

[0037] S102. Perform infrared spectral analysis on light scattering intensity and infrared radiation characteristics to obtain spectral data. Extract the characteristic peaks of the target band from the preprocessed spectral data, calculate the peak shift and intensity change caused by changes in environmental parameters, and obtain the characteristic shift under changes in environmental parameters.

[0038] Fourier transform is used to perform frequency domain conversion on the light scattering intensity and infrared radiation characteristics. The spectral distribution of the time-domain signal is calculated using a fast Fourier transform algorithm to obtain spectral data containing amplitude and phase information. An original spectral curve is generated based on the amplitude distribution of each frequency component in the amplitude information. For the original spectral curve, a moving average filter is used to denoise the spectral data. By setting the filter window width, high-frequency noise components are removed to obtain a smooth spectral curve. Then, a polynomial fitting method is used to perform baseline fitting on the smooth spectral curve. The fitted baseline is subtracted from the smooth spectral curve to obtain the baseline-corrected spectral curve. From the baseline-corrected spectral curve, local maxima within the target band are identified using the second derivative method. If the intensity of a maximum point is higher than a preset threshold and the wavelength interval between it and adjacent extreme points is greater than a resolution limit, it is determined to be a characteristic peak. The wavelength position and peak intensity of the characteristic peak are recorded, along with the currently measured environmental parameter values. Based on the wavelength position, peak intensity, and environmental parameter values ​​of the characteristic peak, the environmental parameters are changed and the spectral measurement and characteristic peak extraction process is repeated to obtain characteristic peak data under different environmental parameters. The difference in wavelength position of the same characteristic peak under different environmental parameters is calculated to obtain the peak offset. The relative rate of change of peak intensity is calculated to obtain the intensity change. The peak offset and intensity change together constitute the characteristic offset under the change of environmental parameters.

[0039] Specifically, the Fourier transform plays a central role in spectral analysis, as it can convert the light scattering intensity and infrared radiation characteristic signals in the time domain to the frequency domain for analysis.

[0040] Specifically, the Fast Fourier Transform (FFT) algorithm decomposes continuously acquired light intensity signals into a superposition of different frequency components, each with specific amplitude and phase information. The amplitude information reflects the contribution of that frequency component to the original signal, while the phase information characterizes its time delay. By extracting the amplitude values ​​of each frequency component from the amplitude information, a raw spectral curve reflecting the material's optical properties can be constructed.

[0041] In one possible implementation, the raw spectral curve often contains various noise interferences, primarily originating from detector thermal noise, environmental vibrations, and electromagnetic interference. Moving average filters suppress high-frequency noise by weighted averaging of adjacent data points; the choice of filter window width requires a balance between denoising effectiveness and spectral resolution. An excessively large window leads to blurred spectral features, while an excessively small window results in poor denoising. Even after filtering, the smoothed spectral curve still suffers from baseline drift, typically caused by nonlinear instrument response and sample scattering background.

[0042] It should be noted that the polynomial fitting method approximates the variation trend of the spectral baseline by selecting a polynomial of appropriate order. During the fitting process, the coefficients of the polynomial are determined using the least squares method, ensuring that the fitted curve achieves the best match with the non-peak regions in the spectral data. Subtracting the fitted baseline from the smoothed spectral curve eliminates the influence of baseline drift, yielding a baseline-corrected spectral curve that truly reflects the sample characteristics. The second derivative method has a unique advantage in characteristic peak identification; it enhances peak characteristics and suppresses background by calculating the second derivative of the spectral curve. When a peak exists in the spectral curve, its second derivative exhibits a negative minimum at the peak position. By finding the local minimum point of the second derivative, the wavelength position of the characteristic peak can be accurately located. The determination of peak intensity needs to consider the signal-to-noise ratio; only when the peak intensity exceeds a certain multiple of the noise level is it considered a true characteristic peak. Recording environmental parameters such as temperature, humidity, and pressure is crucial for subsequent analysis.

[0043] Preferably, changes in environmental parameters cause alterations in the optical properties of the material, leading to a shift in characteristic peaks. Increased temperature alters the intermolecular distance due to thermal expansion, causing characteristic peaks to shift towards longer wavelengths; increased humidity affects the adsorption state on the material surface, changing the intensity of characteristic peaks. By systematically changing environmental parameters and repeating spectral measurements, the variation of characteristic peaks with environmental conditions can be obtained. The peak shift is calculated by determining the wavelength difference of the same characteristic peak under different environmental conditions, while the intensity change is determined by calculating the relative percentage change in peak intensity.

[0044] S103. For the feature offset, construct a dynamic association knowledge graph of surface characteristics of fly ash and steel slag, train a multi-node association model, input the dynamic change trend of feature offset captured by time series analysis method into the multi-node association model, and output the nonlinear interaction relationship between humidity, temperature and surface characteristics.

[0045] For the aforementioned feature offsets and their corresponding environmental parameter data, nodes and edges of a knowledge graph are constructed. These nodes include feature offset nodes, humidity nodes, and temperature nodes, as well as fly ash surface characteristic nodes and steel slag surface characteristic nodes extracted from historical data. The weight values ​​of edges are determined by calculating the Pearson correlation coefficient between the data of each node. If the absolute value of the correlation coefficient is greater than a preset threshold, a connection edge is established between the corresponding nodes, forming a dynamic association knowledge graph of surface characteristics. Based on this dynamic association knowledge graph, a graph convolutional network is used to aggregate node features. The connection relationship between nodes is defined through an adjacency matrix. The feature vector of each node is weighted and summed with the feature vectors of its neighboring nodes. After processing with an activation function, the node representation is updated. Iterative aggregation and update operations are performed to obtain node embeddings containing neighborhood information. Based on the node embeddings and historical feature offset data, a sliding window method is used to segment the time series data. The mean and rate of change of feature offsets within each time window are calculated. The node embeddings are input as state vectors into a long short-term memory network to capture the dynamic trend of feature offset changes, outputting a time-dependent feature offset prediction sequence. Based on the predicted feature offset sequence and node embedding, a multilayer perceptron is constructed to concatenate the humidity node embedding, temperature node embedding, and feature offset node embedding. An interaction feature vector is obtained by mapping through a nonlinear activation function. The components in the interaction feature vector are analyzed to obtain the quantized values ​​of the main effect of humidity, the main effect of temperature, and the interaction effect of humidity and temperature. The nonlinear interaction relationship between humidity, temperature, and surface properties is output.

[0046] Specifically, the construction of knowledge graphs is fundamental to understanding the relationship between material surface properties and environmental parameters.

[0047] Specifically, the feature offset node stores the offset data of spectral features as the environment changes, the humidity node records the temporal variation of relative humidity, and the temperature node contains dynamic monitoring data of ambient temperature. The fly ash surface characteristic node covers multi-dimensional features such as porosity, specific surface area, and particle size distribution, while the steel slag surface characteristic node includes attributes such as oxide layer thickness, surface roughness, and chemical composition distribution. The calculation of the Pearson correlation coefficient quantifies the degree of linear correlation between these nodes. When the absolute value of the correlation coefficient exceeds 0.7, it indicates a strong correlation between the nodes. In this case, connecting edges are established in the graph, and the weight of the edge is the correlation coefficient value.

[0048] In one possible implementation, graph convolutional networks propagate and update node features through a neighborhood aggregation mechanism. The adjacency matrix defines the graph's topology, with elements indicating the existence and strength of connections between nodes. During aggregation, each node collects feature information from its neighbors and integrates this information through a weighted sum. The weights are determined by the adjacency matrix and learnable parameters. The activation function introduces a non-linear transformation, enabling the network to learn complex feature representations. After multiple iterative updates, the embedding of each node not only contains its own feature information but also incorporates information from its multi-hop neighbors, forming a node representation rich in graph structure information.

[0049] It's important to note that the sliding window method plays a crucial role in time series analysis. The choice of window size requires a balance between temporal resolution and statistical stability; smaller windows can capture rapid changes, while larger windows provide more stable statistical properties. The mean of the feature offsets within each window reflects the average level of that period, while the rate of change is calculated by differencing adjacent time points, reflecting the speed and direction of change. Long Short-Term Memory (LSTM) networks are particularly well-suited for processing this type of time series data. Their internal forgetting, input, and output gate mechanisms can selectively retain or forget historical information, effectively solving the gradient vanishing problem of traditional recurrent neural networks. Multilayer perceptrons (MLPs) demonstrate powerful capabilities in handling nonlinear mappings. Humidity node embedding, temperature node embedding, and feature offset node embedding are concatenated to form a high-dimensional input vector. This concatenation preserves the independence of each feature while providing a foundation for subsequent interaction modeling. Nonlinear activation functions such as the hyperbolic tangent or rectified linear units can introduce nonlinear transformations, enabling the network to learn complex interaction patterns between input features. The main effect of humidity reflects the influence of humidity changes alone on surface properties, the main effect of temperature reflects the result of temperature acting independently, while the interaction effect of humidity and temperature reveals the synergistic or antagonistic effects produced when the two act together. This decomposition clearly demonstrates the nonlinear interaction between environmental parameters and material surface properties, providing a theoretical basis for predicting material performance under complex environments.

[0050] S104. Determine the correlation between environmental parameters and the surface properties of fly ash and steel slag based on the nonlinear interaction between humidity, temperature and surface properties.

[0051] Based on the nonlinear interaction between humidity, temperature, and surface properties, the main effect values ​​of humidity, temperature, and the interaction effect value of humidity and temperature are extracted. Normalization is applied to ensure all effect values ​​are within the same order of magnitude. The contribution ratio of each effect to the change in surface properties is calculated. If the interaction effect value of humidity and temperature exceeds a preset threshold, it is determined to be a strongly coupled state, resulting in an environmental parameter coupling strength index. Based on the environmental parameter coupling strength index and the surface property data from the nonlinear interaction, a three-dimensional tensor is constructed. The first dimension is the humidity interval index, the second dimension is the temperature interval index, and the third dimension is the surface property response value. Singular value decomposition (SVD) is used to decompose the tensor, obtaining feature vectors representing the main change patterns. The influence patterns of different environmental condition combinations on surface properties are determined based on these feature vectors. According to the response value distribution of each environmental condition combination in the influence patterns, the kernel density estimation method is used to calculate the probability density function of surface properties under each environmental parameter combination. The expected value and variance of the probability density function are calculated through numerical integration. Based on the variance, the environmental parameter space is divided into a stable response region, a transitional response region, and a sensitive response region, generating an environmental parameter partitioning mapping relationship. Based on the environmental parameter zoning mapping relationship, the average value of the response value in each zoning is calculated as the representative value of that zoning. The frequency of each zoning in actual measurement is counted as the frequency weight. The representative value and the frequency weight are multiplied and summed to obtain the comprehensive response index. The correlation representation between environmental parameters and surface characteristics is constructed, and the correlation representation between environmental parameters and surface characteristics of fly ash and steel slag is determined.

[0052] Specifically, the decomposition of effects of nonlinear interactions is a key step in understanding the influence of environmental parameters.

[0053] It should be noted that the main effect value of humidity reflects the degree of independent influence of humidity changes on the surface properties of materials when the temperature remains constant. When the ambient humidity increases from 30% to 80%, more water molecules are adsorbed on the surface of fly ash particles, leading to enhanced capillary action between particles. This change is directly reflected in the alteration of light scattering characteristics. The main effect value of temperature, on the other hand, characterizes the changes in material properties caused by temperature changes under constant humidity conditions. Increased temperature accelerates the formation rate of the oxide layer on the surface of steel slag and also affects the migration of moisture within fly ash.

[0054] In one possible implementation, the humidity-temperature interaction effect reveals the additional impact of the synergistic effect of the two environmental parameters. Under high temperature and humidity conditions, the evaporation and condensation processes of moisture are intensified, leading to complex mass and heat transfer coupling phenomena. Normalization maps effect values ​​of different dimensions to a unified range of 0 to 1, making the contribution ratios of each effect comparable. When the interaction effect value exceeds 0.3, it indicates a significant synergistic effect between the environmental parameters; at this point, considering the influence of each parameter individually is insufficient to accurately describe the material's response characteristics.

[0055] Specifically, the construction of the three-dimensional tensor requires discretizing continuous environmental parameters. Humidity is divided into several intervals, each representing a different humidity state; temperature is similarly divided into intervals. The surface characteristic response value corresponding to each humidity and temperature combination is filled into the corresponding position in the tensor. Singular value decomposition (SVD) decomposes this three-dimensional tensor into the product of three matrices, where the magnitude of the singular values ​​reflects the importance of the corresponding patterns. The eigenvectors corresponding to the largest singular values ​​describe the main patterns of environmental parameter changes, which may include linear, step, and periodic patterns. Kernel density estimation estimates the probability density by placing a kernel function at each observation point and then superimposing all kernel functions. Gaussian kernels are commonly used, and their bandwidth parameter determines the smoothness of the estimate. The expected value obtained through numerical integration represents the typical value of the surface characteristics under these environmental conditions, while the variance reflects the stability of the response. Regions with variance less than a threshold are classified as stable response regions, where material properties are insensitive to environmental changes; regions with larger variances are sensitive response regions, requiring focused environmental control.

[0056] Preferably, the construction of the correlation representation comprehensively considers both response intensity and frequency of occurrence. Response intensity is represented by the magnitude of the representative values ​​of each partition, while the frequency weight reflects the probability of each environmental condition occurring in actual applications. Combining the two yields a comprehensive evaluation index that more closely reflects real-world application scenarios. This correlation representation not only includes the influence of environmental parameters on material surface properties but also considers the distribution characteristics of actual environmental conditions, providing a quantitative basis for predicting and optimizing material performance under complex environments.

[0057] S105. Extract classification feature semantics from the correlation representation between environmental parameters and surface properties of fly ash and steel slag, and construct a semantically enhanced classification feature set for fly ash and steel slag.

[0058] From the correlation representation of the environmental parameters and the surface characteristics of fly ash and steel slag, each element value of the correlation representation matrix is ​​analyzed, and the numerical correlation strength is converted into descriptive feature words. If the correlation value is greater than a high threshold, it is marked as a "strongly correlated" feature word; if it is between the high and low thresholds, it is marked as a "mediumly correlated" feature word; and if it is below the low threshold, it is marked as a "weakly correlated" feature word. Initial classification feature semantics are formed by combining feature words. Based on the initial classification feature semantics, each parameter combination in the correlation representation is used as a node in the knowledge graph to construct a semantic vector representation, where each feature word corresponds to a dimension. The frequency of feature words appearing in each node is used as a weight value to map the nodes to the semantic vector space, and the cosine similarity between nodes is calculated to obtain the node semantic similarity matrix. According to the node semantic similarity matrix, a threshold-based semantic segmentation method is used to divide the nodes. Multiple similarity thresholds are set to form segmentation boundaries. Nodes with similarity within the same threshold range are grouped into a semantic category. The feature word with the highest frequency in each category is extracted as the semantic label of that category. Based on the semantic categories and semantic tags, node categories containing fly ash features are identified to form a fly ash feature subset, and node categories containing steel slag features are identified to form a steel slag feature subset. By adding high-frequency feature words unique to each subset and basic feature words common to all subsets to each subset, the integrity of semantic expression is enhanced, and semantically enhanced classification feature sets for fly ash and steel slag are output.

[0059] Specifically, the semantic transformation process of the association representation matrix embodies a mapping mechanism from numerical values ​​to language.

[0060] Specifically, a correlation value of 0.85 indicates a very strong response relationship between environmental parameters and material surface properties, thus converting it into a "strongly correlated" feature term. The water absorption and expansion characteristics of fly ash in high humidity environments, and the rapid growth of steel slag oxide layers at high temperatures, both generate high correlation values ​​and are therefore marked as strongly correlated features.

[0061] In one possible implementation, the construction of knowledge graph nodes fully utilizes the structured information in the relational representation. Each parameter combination node contains specific environmental conditions and corresponding material response characteristics.

[0062] For example, "humidity 60% - temperature 80℃ - fly ash porosity increases" constitutes a complete node. In the process of constructing semantic vectors, the choice of feature word dimensions is crucial. "Strongly correlated," "moderately correlated," and so on...

[0063] The basic feature words such as "weak correlation" combined with professional feature words such as "high temperature response", "humidity sensitivity" and "accelerated oxidation" together constitute a multidimensional semantic space.

[0064] It should be noted that using frequency statistics to calculate feature word weights has practical significance. When "

[0065] When strongly correlated feature words appear repeatedly in multiple nodes, it indicates a universal feature, and their weight decreases accordingly. Conversely, certain special feature words, such as "critical humidity mutation," appear only in a few nodes and thus receive higher weights because they represent unique physical phenomena. Cosine similarity calculation effectively measures the semantic proximity between nodes; the smaller the angle between the vectors of two nodes, the more similar they are in the semantic space. Threshold-based semantic segmentation methods discretize the continuous similarity space. Multiple thresholds, such as 0.8, 0.6, and 0.4, are set to divide the similarity space into different intervals. Node pairs with a similarity greater than 0.8 are considered to belong to the same semantic category, often representing similar environmental response patterns. By extracting high-frequency feature words as category labels, the core features of each category can be intuitively understood.

[0066] Preferably, the semantic enhancement process improves classification accuracy through refined management of feature words. The fly ash feature subset includes unique feature words such as "porosity variation," "water absorption characteristics," and "particle dispersion," which accurately describe the physicochemical behavior of fly ash. The steel slag feature subset emphasizes features such as "oxide layer thickness," "surface hardening," and "chemical stability." Common basic feature words such as "temperature dependence" and "humidity influence" reflect the shared environmental response characteristics of the two materials. Through this hierarchical semantic structure, not only is the uniqueness of each material preserved, but their similarities in environmental response are also reflected, forming a complete and hierarchical classification feature set.

[0067] S106. Based on the nonlinear response characteristics of fly ash and steel slag extracted from the classification feature set, optimize the recognition accuracy of the classification features to obtain the optimized classification feature set, classify solid waste types, and generate a preliminary classification model.

[0068] Based on the environmental response data in the classification feature set, the rate of change of fly ash feature offset over time and the nonlinear rate of change of steel slag surface characteristic values ​​are extracted as nonlinear response characteristics. The discriminative power is determined by calculating the overlap of the numerical distribution of the same feature in fly ash and steel slag categories. Redundant features with overlap exceeding a preset threshold are removed, and features with high discriminative power are retained, resulting in a simplified feature set. Based on the simplified feature set, the information gain method is used to calculate the contribution of each feature to solid waste type classification. The feature weights are adjusted according to the contribution magnitude, and the weights are normalized and multiplied by the original feature values ​​to obtain a weighted feature vector. An optimized classification feature set is formed by reconstructing the feature vector. For the optimized classification feature set, a random forest algorithm is used to construct multiple decision trees. Each decision tree randomly selects a subset of features from the feature set for node splitting. The classification results of each decision tree are combined through a voting mechanism. The proportion of decision trees in which the same sample is assigned to the same category is calculated as the consistency ratio. If the consistency ratio exceeds the stability threshold, the classification stability requirement is met. Based on the classification results and stability determination, the decision tree set that meets the stability requirements is integrated to record the classification boundaries and feature thresholds of fly ash and steel slag. Samples whose feature values ​​are in the intersection of the two class boundaries are defined as undetermined classes. A classification rule library containing the determination rules for each class is constructed, and a preliminary classification model for solid waste types is output.

[0069] Specifically, the extraction process of nonlinear response characteristics reveals the dynamic behavior of solid waste materials under environmental changes.

[0070] Specifically, the rate of change of fly ash characteristic offset reflects the speed at which its optical properties evolve over time. When the ambient humidity gradually increases from 40% to 70%, the characteristic offset may initially show a slow increase, but then change sharply when approaching saturation humidity. This nonlinear characteristic is precisely the key indicator for distinguishing different types of solid waste. The nonlinear rate of change of steel slag surface characteristic values ​​is reflected in the stage characteristics of oxide layer growth. The initial oxidation rate is fast, and as the oxide layer thickens, the oxidation rate gradually decreases, exhibiting a typical logarithmic growth pattern.

[0071] In one possible implementation, the discrimination ability is calculated based on the statistical analysis of eigenvalue distributions. For a given feature, its numerical distribution in fly ash and steel slag is statistically analyzed separately, and the discrimination ability is quantified by calculating the overlap area between the two distributions. The smaller the overlap, the stronger the feature's ability to distinguish between the two types of solid waste.

[0072] For example, porosity is generally higher in fly ash, ranging from 0.3 to 0.6, while steel slag has lower porosity, mainly ranging from 0.1 to 0.3. The overlap between the two is very small, so porosity is an excellent distinguishing feature.

[0073] It's important to note that the information gain method assesses a feature's importance by measuring how much it reduces classification uncertainty. In solid waste classification, if a feature can clearly separate a mixed fly ash and steel slag sample into two subsets with higher purity, then that feature has a significant information gain. The feature weight adjustment process considers the dimensional differences between different features, and normalization ensures all features are compared on the same scale. The weighted feature vector more accurately reflects the actual contribution of each feature to the classification task. The core of the random forest algorithm lies in improving the stability and accuracy of classification by integrating multiple decision trees. Each decision tree is constructed using different feature subsets and training sample subsets; this randomness gives each decision tree a certain degree of diversity. When classifying a new sample, each tree makes its own judgment, and the sample category is ultimately determined through a voting mechanism. The consistency ratio reflects the reliability of the classification results; when more than 80% of the decision trees classify a sample as belonging to the same category, the classification result has a high degree of confidence.

[0074] Preferably, the construction process of the classification rule base fully considers the complexities in practical applications. The rules for classifying fly ash may include: porosity greater than 0.3, infrared characteristic peaks located in a specific wavelength band, and a rapid increase in humidity response. Rules for steel slag emphasize: surface oxide layer thickness exceeding a threshold, a unimodal light scattering intensity distribution, and a slow linear increase in temperature response. For samples whose characteristic values ​​fall within the boundary between two classes, defining them as undetermined is a cautious approach. These samples may be mixtures of two materials or solid waste with a special processing history, requiring further analysis for accurate classification.

[0075] S107. Obtain the classification error samples from the preliminary classification model, iteratively optimize the preliminary classification model to obtain the final classification model, and output the solid waste classification results that adapt to environmental changes.

[0076] From the classification results of the preliminary classification model, misclassified fly ash and steel slag samples are extracted. The feature vector and corresponding environmental parameter data for each misclassified sample are recorded, including temperature, humidity, and feature offset at the time of classification. The distribution patterns of misclassified samples under different environmental conditions are statistically analyzed to obtain a set of misclassified samples and their associated environmental data. Based on this set of misclassified samples and their associated environmental data, a gradient boosting decision tree algorithm is used to optimize the preliminary classification model. The difference between the predicted value and the true category of each misclassified sample is calculated as the residual. A new decision tree is constructed based on the magnitude and direction of the residual. Decision trees are iteratively added, and the contribution of each tree is controlled by the learning rate until the classification error rate is lower than a preset threshold, resulting in the optimized final classification model. For the final classification model, real-time environmental parameter data, including current temperature, humidity, and feature offset obtained through spectral analysis, is acquired. This real-time data is input into the final classification model to determine the solid waste type. Simultaneously, the predicted probability of each category is calculated as a classification reliability index. If the highest probability is lower than a preset reliability threshold, the sample is marked as a sample to be updated. Based on the classification results and the samples to be updated, the node weights in the knowledge graph are adjusted using an online gradient descent method. The association strength values ​​between environmental parameters and surface characteristics are updated according to the new classification results. The semantic similarity between nodes is recalculated and the edge connection relationship is adjusted. The dynamic association knowledge graph of fly ash and steel slag surface characteristics is updated, and the solid waste classification results adapted to the current environmental changes are output.

[0077] Specifically, the analysis of erroneous samples revealed the weaknesses of the preliminary classification model and directions for improvement.

[0078] Specifically, when a sample that is actually fly ash is misclassified as steel slag, by tracing back its feature vector, it can be found that the sample may be under special environmental conditions.

[0079] For example, in extremely high humidity environments, the light scattering characteristics of fly ash, after absorbing a large amount of moisture, may approach the characteristic range of steel slag. Recording complete environmental parameters of these erroneous samples, including temperature (85℃), relative humidity (92%), and abnormally increased feature offset, can help identify key factors leading to classification errors.

[0080] In one possible implementation, the gradient boosting decision tree algorithm improves model performance through incremental adjustments. The residuals reflect the current model's prediction bias; for a fly ash sample misclassified as steel slag, the residual is -1, indicating a need for adjustment towards the fly ash category. Newly constructed decision trees are specifically designed to learn from these high-residual samples, identifying classification rules by analyzing their common features. The learning rate balances convergence speed and stability; a smaller learning rate, such as 0.1, ensures that each new tree contributes only a portion of the correction, avoiding overfitting. Through multiple iterations, the model gradually learns classification rules under extreme environmental conditions.

[0081] It's important to note that acquiring and processing real-time environmental parameters is fundamental to dynamic classification. Temperature sensors continuously monitor changes in ambient temperature, humidity sensors record the moisture content in the air, and feature offsets are obtained through real-time spectral analysis. When the spectrometer detects a shift of 15 wavenumber units in the position of a characteristic peak relative to the standard position, this offset data is immediately recorded and input into the classification model. Classification confidence is calculated based on the class probability distribution output by the model. If the probability of the fly ash category is 0.6, the probability of the steel slag category is 0.4, and the highest probability of 0.6 is lower than the set threshold of 0.8, then the sample is marked as awaiting updating. An online gradient descent method implements a real-time update mechanism for the knowledge graph. When a new classification result is generated, the algorithm calculates the gradient between the current graph parameters and the ideal parameters, adjusting the node weights in the opposite direction of the gradient.

[0082] For example, if the number of times fly ash is correctly classified increases under high temperature and high humidity conditions, the weight of the edge "high temperature and high humidity - enhanced fly ash characteristics" will increase accordingly. The update of the correlation strength value reflects the dynamic change in the relationship between environmental parameters and material properties; a previously weak correlation may become stronger due to the addition of new data.

[0083] Preferably, the recalculation of semantic similarity ensures the structural rationality of the knowledge graph. When the features of some nodes change significantly, their similarity relationships with other nodes also need to be adjusted accordingly. Through the updated similarity matrix, nodes that originally belonged to the same category may be reclassified to form a classification structure that better reflects the characteristics of the current environment. This dynamic update mechanism enables the entire classification system to adapt to continuous environmental changes and maintain a high classification accuracy even under seasonal changes and sudden weather changes.

[0084] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the protection scope of one or more embodiments of this specification.

Claims

1. A smart classification method for industrial solid waste based on big data, characterized in that, The method includes: Data on fly ash porosity and steel slag oxide layer were obtained, surface microstructures were analyzed, and light scattering intensity and infrared radiation characteristics were extracted. Infrared spectral analysis is performed on light scattering intensity and infrared radiation characteristics to obtain spectral data. Characteristic peaks of the target band are extracted from the preprocessed spectral data. Peak shifts and intensity changes caused by changes in environmental parameters are calculated to obtain the characteristic shift under environmental parameter changes. To address feature offsets, a dynamic association knowledge graph of fly ash and steel slag surface characteristics is constructed. The nodes of this knowledge graph include feature offset nodes, humidity nodes, temperature nodes, fly ash surface characteristic nodes extracted from historical data, and steel slag surface characteristic nodes. The weights of the connections between nodes are determined using the Pearson correlation coefficient. If the absolute value of the correlation coefficient is greater than a preset threshold, a connection is established between the corresponding nodes. A multi-node association model is trained using this dynamic association knowledge graph as the data and structural basis. A graph convolutional network is used to aggregate the node features of the knowledge graph and update the node representations to obtain node embeddings. A sliding window method is used to segment the time series data, capturing the dynamic trend of feature offsets. The node embeddings and the dynamic trend are input into a long short-term memory network, outputting a feature offset prediction sequence. A multilayer perceptron is used to concatenate the node embeddings and the feature offset prediction sequence, outputting the nonlinear interaction relationship between humidity, temperature, and surface characteristics. The correlation between environmental parameters and the surface properties of fly ash and steel slag is determined based on the nonlinear interaction between humidity, temperature and surface properties. Classification feature semantics are extracted from the correlation representation between environmental parameters and surface properties of fly ash and steel slag, and a semantically enhanced classification feature set for fly ash and steel slag is constructed. Based on the nonlinear response characteristics of fly ash and steel slag extracted from the classification feature set, the recognition accuracy of the classification features is optimized to obtain an optimized classification feature set, which is used to classify solid waste types and generate a preliminary classification model. The model obtains misclassified samples from the preliminary classification model, iteratively optimizes the preliminary classification model, obtains the final classification model, and outputs solid waste classification results that adapt to environmental changes.

2. The intelligent classification method for industrial solid waste based on big data according to claim 1, characterized in that, The process of acquiring fly ash porosity and steel slag oxide layer data, analyzing surface microstructures, and extracting light scattering intensity and infrared radiation characteristics includes: Two-dimensional cross-sectional images of fly ash particles are acquired to identify pore regions and solid regions, and the porosity is obtained by calculating the ratio of pore area to total area. Energy dispersive spectroscopy (EDS) analysis is performed on the surface of steel slag samples to determine the oxide layer thickness based on the oxygen content distribution. The surfaces of fly ash and steel slag samples are irradiated with a laser, and scattered light signals are collected using a photodetector array to record the intensity of scattered light at each angle. The infrared radiation intensity value is calculated based on the correspondence between the absorption peak intensity and the oxide layer thickness. The maximum intensity value within a preset scattering angle is extracted as the light scattering intensity feature, and the average radiation intensity value within a preset wavelength range in the infrared absorption spectrum is extracted as the infrared radiation feature.

3. The intelligent classification method for industrial solid waste based on big data according to claim 1, characterized in that, The process involves infrared spectral analysis of light scattering intensity and infrared radiation characteristics to obtain spectral data. Characteristic peak values ​​of the target band are extracted from the preprocessed spectral data. Peak shifts and intensity changes caused by environmental parameter variations are calculated to obtain the characteristic shift under environmental parameter changes, including: The light scattering intensity and infrared radiation characteristics are frequency domain transformed to obtain spectral data; the spectral data are denoised using a moving average filter, and baseline correction is performed using a polynomial fitting method to obtain a baseline-corrected spectral curve; from the baseline-corrected spectral curve, characteristic peaks in the target band are identified using the second derivative method, and the wavelength position and peak intensity of the characteristic peaks are recorded.

4. The intelligent classification method for industrial solid waste based on big data according to claim 1, characterized in that, The method for determining the correlation between environmental parameters and the surface properties of fly ash and steel slag based on the nonlinear interaction between humidity, temperature, and surface properties includes: The main effect values ​​of humidity, temperature, and humidity-temperature interaction are extracted from the nonlinear interaction relationship. The contribution ratio of each effect to the surface property change is calculated through normalization to obtain the environmental parameter coupling strength index. A three-dimensional tensor is constructed, and the tensor is decomposed to obtain feature vectors, determining the influence pattern of environmental condition combinations on surface properties. The probability density function of the influence pattern is calculated using the kernel density estimation method, dividing the environmental parameter space into different response zones and generating an environmental parameter partition mapping relationship. The representative value and frequency weight of each partition in the partition mapping relationship are calculated to obtain a comprehensive response index, and the associated representation is constructed.

5. The intelligent classification method for industrial solid waste based on big data according to claim 1, characterized in that, The process involves extracting classification feature semantics from the correlation representation between environmental parameters and the surface properties of fly ash and steel slag, and constructing a semantically enhanced classification feature set for fly ash and steel slag, including: The element values ​​of the association representation are parsed and converted into descriptive feature words to generate initial classification feature semantics; the parameter combination of the association representation is mapped into a semantic vector, and the cosine similarity between nodes is calculated to obtain a semantic similarity matrix; the nodes in the semantic similarity matrix are divided by a threshold-based semantic segmentation method to generate semantic categories and labels; the node categories containing fly ash and steel slag features are identified to form a feature subset, high-frequency feature words and basic feature words are added, and the classification feature set is output.

6. The intelligent classification method for industrial solid waste based on big data according to claim 1, characterized in that, The process involves optimizing the identification accuracy of classification features based on the nonlinear response characteristics of fly ash and steel slag extracted from the classification feature set, obtaining an optimized classification feature set, classifying solid waste types, and generating a preliminary classification model, including: Extract the nonlinear response characteristics from the classification feature set, calculate the feature discrimination, remove redundant features, and obtain a simplified feature set; adjust the feature weights of the simplified feature set using the information gain method to generate a weighted feature vector, and construct an optimized classification feature set; construct a decision tree using the random forest algorithm, calculate the classification consistency ratio, determine stability, and output a preliminary classification model containing classification rules.

7. The intelligent classification method for industrial solid waste based on big data according to claim 1, characterized in that, The process of obtaining misclassified samples from the preliminary classification model, iteratively optimizing the preliminary classification model to obtain the final classification model, and outputting solid waste classification results adapted to environmental changes includes: Extract the erroneous samples and their environmental parameter data from the preliminary classification model to generate an erroneous sample set; calculate the residuals using the gradient boosting decision tree algorithm, iteratively optimize the preliminary classification model, and obtain the final classification model; input real-time environmental parameters, calculate the classification confidence index, and mark samples to be updated; adjust the node weights of the knowledge graph using the online gradient descent method, update the association strength values ​​and edge connections, and output the solid waste classification results.

Images