A multi-parameter adaptive adjustment industrial classification system based on environment perception

Abstract

An environmentally-aware, multi-parameter adaptive adjustment industrial classification system includes an environmental sensing module, a multispectral imaging technology module, and an ultra-high-speed imaging system module located in the sorting workshop, as well as a central control unit located in the control room. A multimodal data fusion algorithm module processes the acquired images of different spectra and multidimensional feature images to extract classification features. The central processing unit incorporates a time-series analysis algorithm to analyze and predict historical data collected by the environmental sensing module. By analyzing the changing trends of environmental parameters, it predicts future changes in environmental parameters and dynamically adjusts imaging parameters to achieve automatic classification of industrial materials. This invention improves classification stability. By monitoring industrial environmental parameters in real time and automatically adjusting imaging technology and algorithm parameters, the classification system can better adapt to complex and changing industrial environments.

Description

Technical Field

[0001] This invention belongs to the field of industrial automation technology, specifically relating to a multi-parameter adaptive adjustment industrial classification system based on environmental perception. Background Technology

[0002] In industrial production, classification systems are one of the key technologies for achieving automated material sorting, quality control, and production management. With the rapid development of industrial automation and intelligence, higher demands are being placed on the accuracy and stability of classification systems. However, traditional classification methods are often based on fixed parameter settings, making them unable to adapt to complex and ever-changing industrial environments, resulting in unstable classification results and impacting production efficiency and product quality.

[0003] Currently, industrial classification systems mainly rely on fixed parameter settings and simple environmental adaptation adjustments. For example, some systems adjust imaging parameters by setting preset light intensity and temperature ranges, but these methods cannot respond to environmental changes in real time, leading to a decline in classification performance when environmental parameters fluctuate significantly. Furthermore, some systems employ simple time-series analysis methods to predict changes in environmental parameters, but due to algorithmic limitations, the prediction accuracy and response speed are insufficient to meet practical requirements.

[0004] Although existing industrial classification systems can adapt to environmental changes to some extent, they still have the following problems and shortcomings: First, existing classification systems lack real-time environmental awareness and cannot dynamically adjust classification parameters according to real-time changes in environmental parameters, resulting in unstable classification performance in complex and ever-changing industrial environments.

[0005] Secondly, existing time series analysis methods have limited prediction accuracy and are difficult to accurately predict long-term trends in environmental parameters, which affects the adaptability and stability of classification systems.

[0006] Furthermore, existing environmental sensors lack sufficient accuracy and sensitivity to accurately monitor minute changes in environmental parameters, preventing classification systems from making precise parameter adjustments. Therefore, developing a multi-parameter adaptive adjustment industrial classification system based on environmental perception has significant practical implications and application value. Summary of the Invention

[0007] The technical problem to be solved by this invention is as follows: 1) Traditional classification methods cannot respond to environmental changes in real time, resulting in unstable classification results in complex and ever-changing industrial environments; 2) Existing time series analysis methods have limited prediction accuracy, making it difficult to accurately predict the long-term changing trends of environmental parameters, which is a technical problem affecting the adaptability and stability of classification systems; 3) The existing environmental sensors are not accurate and sensitive enough to accurately monitor minute changes in environmental parameters, which leads to the classification system being unable to make precise parameter adjustments.

[0008] To solve the above problems, the present invention is achieved through the following technical solution: An industrial classification system based on environmental perception and multi-parameter adaptive adjustment is characterized by the following steps: it includes an environmental perception module, a multispectral imaging technology module, and an ultra-high-speed imaging system module located in the sorting workshop, and a central control unit located in the control room; the central control unit is connected to the environmental perception module, the multispectral imaging technology module, and the ultra-high-speed imaging system module respectively. Its working method includes the following steps: Step 1: The environmental sensing module monitors industrial environmental parameters in real time, including temperature, humidity, and light intensity; Step 2: The central processing unit receives industrial environment parameter data and uses an adaptive band selection algorithm to dynamically select the most suitable spectral band for imaging based on the characteristics of different industrial environments and sample properties. During the imaging process, the parameters of the multispectral imaging technology module and the ultra-high-speed imaging system module are adjusted using a real-time feedback mechanism. Step 3: The multimodal data fusion algorithm module processes the acquired images of different spectra and multidimensional feature images to extract classification features; Step 4: The central processing unit introduces a time series analysis algorithm to analyze and predict the historical data collected by the environmental perception module. By analyzing the changing trends of environmental parameters, it can predict the changes in environmental parameters in the future and dynamically adjust the imaging parameters. Step 5: The central processing unit outputs the processed data to achieve automatic classification of industrial materials.

[0009] The environmental sensing module includes a temperature sensor, a humidity sensor, and a light sensor located in the sorting workshop, which collect temperature, humidity, and light intensity data in the industrial environment, respectively. The data acquisition unit transmits the data collected by the sensors to the central control unit.

[0010] The multispectral imaging technology module includes a support frame located above the sorting assembly line, on which a multispectral camera and a multi-wavelength light source are mounted. Both the multispectral camera and the multi-wavelength light source are connected to a control circuit.

[0011] The ultra-high-speed imaging system module includes a high-speed camera, which includes a wide-angle camera and a microscope camera with their optical paths arranged in parallel. Both the wide-angle camera and the microscope camera are connected to the camera control circuit.

[0012] Step 2 specifically includes: Step 2.1: When the difference between the industrial environmental parameter data and the preset threshold is less than ΔT, the central processing unit selects a light source with a preset wavelength according to the sorted items. In this way, the sorted items are illuminated by a multi-wavelength light source, and the multi-spectral camera in the multi-spectral imaging technology module performs imaging to obtain images of different spectra. Step 2.2: If, within a certain time period, the difference between the industrial environmental parameter data and the preset threshold remains less than ΔT, the central processing unit controls the ultra-high-speed imaging system module to take pictures. The imaging adopts a pyramid-shaped multi-scale, multi-resolution imaging strategy, using different imaging devices and parameters at different scales and resolutions; specifically: Step 2.2.1: Use a low-resolution wide-angle camera to perform global imaging at a large scale to obtain the overall feature information of the sample; Step 2.2.2: Use a high-resolution microscope camera to perform local imaging at a small scale to obtain detailed feature information of the sample; Step 2.2.3: Fuse images at different scales and resolutions to obtain richer and more accurate multidimensional feature images.

[0013] Step 3 specifically includes: Step 3.1: Data preprocessing and registration; Step 3.2: Feature Extraction: (1) Spectral image: PCA or ICA is used for dimensionality reduction to extract the spectral feature vector of the sorted items; (2) Multidimensional feature images: Extract multi-scale edge or direction features using wavelet transform; Step 3.3: Multimodal feature fusion: In the deep learning network, different layers are gradually fused, that is, multiple modalities are input → each goes through several layers of convolution to extract mid-level features → feature maps are concatenated or added together and fused → then passed through high-level networks → final classification; Step 3.4: Feature Selection and Dimensionality Reduction: First, calculate the mutual information between each feature and the category, retaining highly correlated features; second, use L1 regularization to allow the model to automatically sparsify the features. Step 3.5: Classification and Feedback Optimization: The extracted features are finally input into the classifier.

[0014] Step 4 specifically includes: Step 4.1: The historical data collected by the environmental sensing module is stored in the database; Step 4.2: The central processing unit calls the time series analysis algorithm to analyze and predict historical data; Step 4.2.1: Preprocess the collected historical environmental sensor data, including normalization and handling of missing values; Step 4.2.2: Input the preprocessed data into the trained LSTM model for prediction; Step 4.2.2.1: Employ multiple time series analysis algorithms: Combine the ARIMA algorithm with the LSTM neural network algorithm. By fusing the two algorithms, the accuracy of analysis and prediction of environmental parameter change trends can be improved. Step 4.2.3: Based on the prediction results of the LSTM model, establish a dynamic parameter adjustment strategy, with different prediction error ranges corresponding to different system parameter adjustment magnitudes; Step 4.3: Based on the prediction results, the central processing unit adjusts the parameters of the multispectral imaging technology module and the ultra-high-speed imaging system module in advance to adapt to the upcoming environmental changes.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. Improve classification stability: By monitoring industrial environmental parameters in real time and automatically adjusting imaging technology and algorithm parameters, this technical solution enables the classification system to better adapt to complex and ever-changing industrial environments and improve the stability of classification results.

[0016] 2. Enhance system adaptability: By introducing time series analysis algorithms and LSTM networks, this technical solution can predict future changes in environmental parameters, adjust system parameters in advance, and enhance the system's adaptability and stability.

[0017] 3. Improved prediction accuracy: By employing high-precision and high-sensitivity environmental sensors and combining them with LSTM networks to process time-series data with long-term dependencies, this technical solution significantly improves the prediction accuracy of environmental parameter changes.

[0018] 4. Enhanced sensor reliability: By regularly calibrating and adding protective devices, this technical solution improves the stability and reliability of the sensor in complex industrial environments, ensuring the accuracy of environmental parameter monitoring.

[0019] 5. Optimize system parameter adjustment: Based on the prediction results, a dynamic parameter adjustment strategy is established. This technical solution can flexibly adjust system parameters according to different prediction error ranges to further improve the classification effect. Attached Figure Description

[0020] Figure 1 This is a flowchart of the present invention; Figure 2 Timing diagram for LSTM environment prediction and parameter tuning; Figure 3 This is a schematic diagram of the principle of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0022] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and 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. Therefore, they should not be construed as limitations on this invention.

[0023] like Figure 1 As shown, an environmental perception-based multi-parameter adaptive adjustment industrial classification system includes an environmental perception module, a multispectral imaging technology module, and an ultra-high-speed imaging system module located in the sorting workshop, as well as a central control unit located in the control room; the central control unit is connected to the environmental perception module, the multispectral imaging technology module, and the ultra-high-speed imaging system module respectively.

[0024] The environmental sensing module includes a temperature sensor, a humidity sensor, and a light sensor located in the sorting workshop, which collect temperature, humidity, and light intensity data in the industrial environment, respectively. The data acquisition unit transmits the data collected by the sensors to the central control unit.

[0025] It should be noted that the sensors in the environmental sensing module can be installed anywhere in the sorting workshop, such as on workshop walls, near the sorting line, around multispectral imaging equipment, or in sample placement areas, with multiple environmental sensing modules distributed in a reasonable manner. This distributed sensor network of multiple environmental sensing modules transmits data from each sensor to the central control unit via wireless communication modules. This allows for comprehensive and accurate acquisition of parameters such as temperature, humidity, and light intensity at different locations in the industrial environment, avoiding monitoring errors caused by local environmental differences. For example, in large industrial workshops, light intensity and temperature may vary significantly in different areas; the distributed sensor network can more accurately reflect the overall environmental conditions.

[0026] The multispectral imaging technology module includes a support frame positioned above the sorting assembly line. A multispectral camera and a multi-wavelength light source are mounted on the support frame, both connected to a control circuit. The multispectral camera is equipped with a filter; the multi-wavelength light source refers to an LED light group or laser that emits different specific wavelengths (such as blue light, green light, red light, near-infrared light, etc.). In industrial or darkroom environments, it actively illuminates the target to eliminate ambient light interference and obtain a stable and controllable spectral response.

[0027] In this way, under illumination by multi-wavelength light sources, the multispectral camera acquires images of different spectra through filters, and the control circuit adjusts the light intensity of the multi-wavelength light source and the exposure time of the multispectral camera according to environmental parameters.

[0028] The ultra-high-speed imaging system module includes a high-speed camera, which comprises a wide-angle camera and a microscope camera arranged in parallel optical paths. Both the wide-angle camera and the microscope camera are connected to the camera control circuit and are also connected to an image memory. Under the control of the camera control circuit, the high-speed camera captures images at a high frame rate, and the image memory stores the captured image data for subsequent processing.

[0029] The central control unit includes a processor, an algorithm library, and corresponding software. The central processing unit receives data from the environmental perception module and the multispectral imaging technology module, and calls the multimodal data fusion algorithm in the algorithm library for processing. The working method of the environmentally-aware, multi-parameter adaptive adjustment industrial classification system includes the following steps: Step 1: The environmental sensing module monitors industrial environmental parameters in real time, including temperature, humidity, and light intensity; Step 2: The central processing unit receives industrial environmental parameter data and uses an adaptive band selection algorithm to dynamically select the most suitable spectral band for imaging based on the characteristics of different industrial environments and sample properties. During the imaging process, a real-time feedback mechanism is used to adjust the parameters of the multispectral imaging technology module and the ultra-high-speed imaging system module to improve image quality and the accuracy of feature information. Simultaneously, a multi-scale, multi-resolution imaging strategy is introduced to acquire multi-dimensional feature information of the sample from different scales and resolutions, increasing the richness of the information. The specific process is as follows: Step 2.1: When the difference between the industrial environmental parameter data and the preset threshold is less than ΔT, the central processing unit selects a light source with a preset wavelength according to the sorted items. In this way, the sorted items are illuminated by a multi-wavelength light source, and the multi-spectral camera in the multi-spectral imaging technology module performs imaging to obtain images of different spectra. Step 2.2: If, within a certain time period, the difference between the industrial environmental parameter data and the preset threshold remains less than ΔT, the central processing unit controls the ultra-high-speed imaging system module to take pictures. The imaging adopts a pyramid-shaped multi-scale, multi-resolution imaging strategy, using different imaging devices and parameters at different scales and resolutions; specifically: Step 2.2.1: Use a low-resolution wide-angle camera to perform global imaging at a large scale to obtain the overall feature information of the sample; Step 2.2.2: Use a high-resolution microscope camera to perform local imaging at a small scale to obtain detailed feature information of the sample; Step 2.2.3: Fuse images at different scales and resolutions to obtain richer and more accurate multidimensional feature images.

[0030] Step 3: The multimodal data fusion algorithm module processes the acquired images of different spectra and multidimensional feature images to extract classification features.

[0031] Step 3.1: Data Preprocessing and Registration: Data acquired by different sensors may have inconsistent resolution, viewing angle, and time. Therefore, images with different spectra and multidimensional feature images are aligned to the same spatial coordinate system; at the same time, outliers are eliminated. Step 3.2: Feature Extraction: (1) Spectral image: PCA or ICA is used for dimensionality reduction to extract the spectral feature vector of the sorted items; (2) Multidimensional feature images: Extract multi-scale edge or direction features using wavelet transform; Step 3.3: Multimodal feature fusion: Different layers in the deep learning network are gradually fused, that is, multiple modalities are input → each goes through several layers of convolution to extract mid-level features → feature maps are concatenated or added together and fused → then go through high-level networks → final classification.

[0032] Step 3.4: Feature Selection and Dimensionality Reduction: First, calculate the mutual information between each feature and the category, retaining highly correlated features; second, use L1 regularization (Lasso) to allow the model to automatically sparsify the features.

[0033] Step 3.5: Classification and Feedback Optimization: The extracted features are finally input into a classifier (SVM, Random Forest, CNN, Transformer, etc.).

[0034] Step 4: The central processing unit (CPU) dynamically adjusts imaging parameters based on changes in environmental parameters to adapt to the rapidly changing industrial environment. The CPU incorporates time-series analysis algorithms to analyze and predict historical data collected by the environmental sensing module. By analyzing trends in environmental parameters, it predicts future changes in environmental parameters, enabling the system to better adapt to upcoming environmental changes and improving classification stability. Figure 2 As shown, the specific process is as follows: Step 4.1: The historical data collected by the environmental sensing module is stored in the database; Step 4.2: The central processing unit calls the time series analysis algorithm to analyze and predict historical data; Step 4.2.1: Preprocess the collected historical environmental sensor data, including normalization and handling of missing values; Step 4.2.2: Input the preprocessed data into the trained LSTM model for prediction; Step 4.2.2.1: Employing Multiple Time Series Analysis Algorithms: Combining the ARIMA algorithm with the LSTM neural network algorithm. The ARIMA algorithm excels at processing linear time series data, while the LSTM neural network algorithm can capture the nonlinear characteristics in time series. By fusing the two algorithms, the accuracy of analyzing and predicting the changing trends of environmental parameters is improved. Simultaneously, an attention mechanism is introduced into the LSTM network, allowing the model to focus more on key information in the time series, further enhancing prediction performance.

[0035] i. First, use ARIMA to capture linear components, then use ARIMA to analyze the original sequence. y t Perform fitting to obtain predicted values ŷ t and residual sequence et = y t - ŷ t ; ii. Then use LSTM to learn the nonlinear patterns in the residuals (i.e., errors) that ARIMA failed to capture: The residual sequence... et The objective is to train an LSTM model; the input to the LSTM can be the residuals from the past few time steps, and the output is the predicted value of the residuals for the next time step. ĕ t+1 ; iii. Final prediction: Final 预测 = ŷ t+1 (From ARIMA) + ĕ t+1 (From LSTM).

[0036] Step 4.2.3: Based on the prediction results of the LSTM model, establish a dynamic parameter adjustment strategy. Different prediction error ranges correspond to different system parameter adjustment ranges.

[0037] A parameter adjustment rule base was established, and detailed parameter adjustment rules were formulated based on different combinations and degrees of change of environmental parameters. When the temperature change is within ±5℃, slight temperature compensation is applied to the multimodal data fusion algorithm; when the temperature change exceeds ±5℃, in addition to compensating the data fusion algorithm, the gain parameters of the multispectral imaging technology are also adjusted. Simultaneously, an adaptive adjustment mechanism is introduced to dynamically optimize the parameter adjustment rules based on the actual classification performance of the system.

[0038] Step 4.3: Based on the prediction results, the central processing unit adjusts the parameters of the multispectral imaging technology module and the ultra-high-speed imaging system module in advance to adapt to the upcoming environmental changes.

[0039] Thus, this invention constructs an intelligent agent that continuously learns and experiments in different industrial environments and sample characteristics to dynamically select the most suitable spectral band. The agent adjusts its decision-making strategy based on reward signals from the environment, gradually finding the optimal band selection scheme. Simultaneously, by combining historical and real-time data, the algorithm is updated and optimized in real time, improving its adaptability and accuracy.

[0040] Step 5: The central processing unit outputs the processed data to achieve automatic classification of industrial materials.

[0041] Through the above technical solutions and embodiments, this invention successfully solves the technical problem that traditional classification methods cannot adapt to complex and ever-changing industrial environments, leading to unstable classification results. By monitoring environmental parameters in real time and dynamically adjusting classification parameters, the accuracy and stability of classification are improved. Simultaneously, by introducing time series analysis algorithms and LSTM models, the prediction accuracy of environmental parameter change trends is improved, further enhancing the adaptability and stability of the classification system.

[0042] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several changes and improvements without departing from the overall concept of the present invention, and these should also be considered within the scope of protection of the present invention.

Claims

1. A multi-parameter adaptive adjustment industrial classification system based on environmental perception, characterized in that: The process includes the following steps: an environmental sensing module, a multispectral imaging technology module, and an ultra-high-speed imaging system module located in the sorting workshop, and a central control unit located in the control room; the central control unit is connected to the environmental sensing module, the multispectral imaging technology module, and the ultra-high-speed imaging system module respectively. Its working method includes the following steps: Step 1: The environmental sensing module monitors industrial environmental parameters in real time, including temperature, humidity, and light intensity; Step 2: The central processing unit receives industrial environment parameter data and uses an adaptive band selection algorithm to dynamically select the most suitable spectral band for imaging based on the characteristics of different industrial environments and sample properties. During the imaging process, the parameters of the multispectral imaging technology module and the ultra-high-speed imaging system module are adjusted using a real-time feedback mechanism. Step 3: The multimodal data fusion algorithm module processes the acquired images of different spectra and multidimensional feature images to extract classification features; Step 4: The central processing unit introduces a time series analysis algorithm to analyze and predict the historical data collected by the environmental perception module. By analyzing the changing trends of environmental parameters, it can predict the changes in environmental parameters in the future and dynamically adjust the imaging parameters. Step 5: The central processing unit outputs the processed data to achieve automatic classification of industrial materials.

2. The multi-parameter adaptive adjustment industrial classification system based on environmental perception according to claim 1, characterized in that: The environmental sensing module includes a temperature sensor, a humidity sensor, and a light sensor located in the sorting workshop, which collect temperature, humidity, and light intensity data in the industrial environment, respectively. The data acquisition unit transmits the data collected by the sensors to the central control unit.

3. The multi-parameter adaptive adjustment industrial classification system based on environmental perception according to claim 1, characterized in that: The multispectral imaging technology module includes a support frame located above the sorting assembly line, on which a multispectral camera and a multi-wavelength light source are mounted. Both the multispectral camera and the multi-wavelength light source are connected to a control circuit.

4. The multi-parameter adaptive adjustment industrial classification system based on environmental perception according to claim 1, characterized in that: The ultra-high-speed imaging system module includes a high-speed camera, which includes a wide-angle camera and a microscope camera with their optical paths arranged in parallel. Both the wide-angle camera and the microscope camera are connected to the camera control circuit.

5. The multi-parameter adaptive adjustment industrial classification system based on environmental perception according to claim 1, characterized in that: Step 2 specifically includes: Step 2.1: When the difference between the industrial environmental parameter data and the preset threshold is less than ΔT, the central processing unit selects a light source with a preset wavelength according to the sorted items. In this way, the sorted items are illuminated by a multi-wavelength light source, and the multi-spectral camera in the multi-spectral imaging technology module performs imaging to obtain images of different spectra. Step 2.2: If, within a certain time period, the difference between the industrial environmental parameter data and the preset threshold remains less than ΔT, the central processing unit controls the ultra-high-speed imaging system module to take pictures. The imaging adopts a pyramid-shaped multi-scale, multi-resolution imaging strategy, using different imaging devices and parameters at different scales and resolutions; specifically: Step 2.2.1: Use a low-resolution wide-angle camera to perform global imaging at a large scale to obtain the overall feature information of the sample; Step 2.2.2: Use a high-resolution microscope camera to perform local imaging at a small scale to obtain detailed feature information of the sample; Step 2.2.3: Fuse images at different scales and resolutions to obtain richer and more accurate multidimensional feature images.

6. The multi-parameter adaptive adjustment industrial classification system based on environmental perception according to claim 1, characterized in that: Step 3 specifically includes: Step 3.1: Data preprocessing and registration; Step 3.2: Feature Extraction: (1) Spectral image: PCA or ICA is used for dimensionality reduction to extract the spectral feature vector of the sorted items; (2) Multidimensional feature images: Extract multi-scale edge or direction features using wavelet transform; Step 3.3: Multimodal feature fusion: In the deep learning network, different layers are gradually fused, that is, multiple modalities are input → each goes through several layers of convolution to extract mid-level features → feature maps are concatenated or added together and fused → then passed through high-level networks → final classification; Step 3.4: Feature Selection and Dimensionality Reduction: First, calculate the mutual information between each feature and the category, retaining highly correlated features; second, use L1 regularization to allow the model to automatically sparsify the features. Step 3.5: Classification and Feedback Optimization: The extracted features are finally input into the classifier.

7. The multi-parameter adaptive adjustment industrial classification system based on environmental perception according to claim 1, characterized in that: Step 4 specifically includes: Step 4.1: The historical data collected by the environmental sensing module is stored in the database; Step 4.2: The central processing unit calls the time series analysis algorithm to analyze and predict historical data; Step 4.2.1: Preprocess the collected historical environmental sensor data, including normalization and handling of missing values; Step 4.2.2: Input the preprocessed data into the trained LSTM model for prediction; Step 4.2.2.1: Employ multiple time series analysis algorithms: Combine the ARIMA algorithm with the LSTM neural network algorithm. By fusing the two algorithms, the accuracy of analysis and prediction of environmental parameter change trends can be improved. Step 4.2.3: Based on the prediction results of the LSTM model, establish a dynamic parameter adjustment strategy, with different prediction error ranges corresponding to different system parameter adjustment magnitudes; Step 4.3: Based on the prediction results, the central processing unit adjusts the parameters of the multispectral imaging technology module and the ultra-high-speed imaging system module in advance to adapt to the upcoming environmental changes.

Images