Online sorting method and system based on a color sorter
By applying relevance feature selection and unsupervised machine learning algorithms combined with image segmentation in color sorters, the color, texture, and geometric features of materials are extracted, solving the problem of insufficient recognition accuracy of traditional color sorters and achieving more efficient material sorting.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEFEI BOYAN SHIXIN ELECTRONIC TECH CO LTD
- Filing Date
- 2023-03-10
- Publication Date
- 2026-06-23
AI Technical Summary
Traditional color sorters rely primarily on color and texture differences in their algorithms, resulting in insufficient recognition accuracy and an inability to guarantee the quality of sorting results.
The correlation feature selection (CFS) algorithm combined with the best first search (BF) algorithm is used for dimensionality reduction. Unsupervised machine learning techniques (LDA) and multilayer perceptron (MLP) are used to classify material feature data. Image segmentation processing algorithms are used to eliminate noise and extract color, texture and geometric features for identification.
It significantly improves the accuracy of material identification and sorting quality, and enhances the sorting effect of color sorters.
Smart Images
Figure CN116351731B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of color sorter technology, and in particular to an online sorting method and system based on a color sorter. Background Technology
[0002] Color sorters are products that combine optics, mechanics, software development, image acquisition, and processing. Due to their high classification accuracy, fast rejection speed, and adaptable sorting algorithms, color sorters are widely used for sorting materials such as grains, tea, and ores. Specifically, color sorters use a conveyor mechanism to send large quantities of materials to be sorted to an image acquisition and rejection device. The image acquisition device, such as a linear CCD, acquires real-time image information of the materials. After image processing and classification according to rejection rules, unqualified materials flow through the rejection device, where an air valve is opened to separate the materials. However, traditional color sorting methods primarily use algorithms based on color and texture differences. These algorithms are not comprehensive enough, and the accuracy of material classification needs improvement. Furthermore, traditional techniques cannot guarantee the quality of the sorting results. Summary of the Invention
[0003] To address the technical problems existing in the background art, this invention proposes an online sorting method and system based on a color sorter.
[0004] The present invention proposes an online sorting method based on a color sorter, comprising the following steps:
[0005] Receive the material to be identified;
[0006] The received materials to be identified are then identified to obtain the identification results;
[0007] Based on the identification results, reject the unqualified materials;
[0008] The process involves identifying the received materials to be identified and obtaining the identification results, specifically including:
[0009] Acquire images of the material to be identified;
[0010] Preprocess the acquired images of the materials to be identified;
[0011] Feature extraction is performed on the preprocessed image of the material to be identified to obtain the feature data of the material to be identified; the feature data of the material to be identified includes color and texture features as well as geometric features;
[0012] The dimensionality reduction and classification of the material feature data to be identified are performed to obtain the identification results.
[0013] Preferably, the geometric features include area, aspect ratio, major axis, minor axis, and equivalent diameter.
[0014] Preferably, the feature data of the material to be identified is subjected to dimensionality reduction and classification to obtain the identification result, specifically including:
[0015] The correlation feature selection (CFS) algorithm combined with the best first search (BF) algorithm is used to reduce the dimensionality of the material feature data to be identified, and the dimensionality reduction result is obtained.
[0016] The unsupervised machine learning technique LDA is used to classify the dimensionality reduction results to obtain the identification results.
[0017] Preferably, the correlation feature selection (CFS) algorithm combined with the best-first search (BF) algorithm is used to reduce the dimensionality of the material feature data to be identified, resulting in the dimensionality reduction result, specifically including:
[0018] The Correlation Feature Selection (CFS) algorithm is used to select features from the material feature data to be identified, and the selected features are used as the dataset.
[0019] Calculate the feature-category and feature-feature correlation matrices from the dataset;
[0020] The feature subset space is obtained by using the Best-First Search (BF) algorithm.
[0021] Calculate the estimated values of feature subsets in the feature subset space, and find the feature subset with the largest estimated value as the optimal feature subset;
[0022] The dimensionality of the feature subset space is reduced by using the optimal feature subset, and the dimensionality reduction result is obtained.
[0023] Preferably, the dimensionality reduction results are classified using the unsupervised machine learning technique LDA to obtain the recognition results, specifically including:
[0024] The LDA algorithm of unsupervised machine learning is used to perform LDA transformation on the dimensionality reduction result to obtain the feature vector;
[0025] The feature vectors are classified using a multilayer perceptron (MLP) to obtain the classification results for each layer.
[0026] The classification results of each layer are integrated to obtain the recognition results.
[0027] Preferably, the acquired images of the material to be identified are preprocessed, specifically including:
[0028] Correct the image of the material to be identified;
[0029] Image segmentation processing algorithms are used to segment the material to be identified from the background in the image of the material to be identified.
[0030] Preferably, the image segmentation processing algorithm includes the optimal thresholding method, the Otsu thresholding method, and the HSI system thresholding method.
[0031] Preferably, after segmenting the material to be identified from the background in the image of the material to be identified using an image segmentation processing algorithm, the method further includes:
[0032] Eliminate noise.
[0033] This invention also proposes an online sorting system based on a color sorter, comprising:
[0034] The feeding device is used to receive the material to be identified;
[0035] An identification device is used to identify received materials to be identified and obtain identification results.
[0036] A rejection device is used to reject unqualified materials based on the identification results;
[0037] The identification device includes:
[0038] The image acquisition module is used to acquire images of the material to be identified.
[0039] The preprocessing module is used to preprocess the acquired images of the materials to be identified.
[0040] The feature extraction module is used to extract features from the preprocessed image of the material to be identified to obtain the feature data of the material to be identified; wherein, the feature data of the material to be identified includes color and texture features as well as geometric features;
[0041] The identification module is used to reduce the dimensionality and classify the feature data of the material to be identified to obtain the identification result.
[0042] Preferably, the geometric features include area, aspect ratio, major axis, minor axis, and equivalent diameter.
[0043] Preferably, the identification module includes:
[0044] The dimensionality reduction submodule is used to perform dimensionality reduction processing on the feature data of the material to be identified by combining the correlation feature selection (CFS) algorithm with the best first search (BF) algorithm to obtain the dimensionality reduction result.
[0045] The recognition submodule is used to classify the dimensionality reduction results using the unsupervised machine learning technique LDA to obtain the recognition results.
[0046] Preferably, the recognition submodule includes an LDA model and an MLP model. The LDA model is used to output a feature vector based on the dimensionality reduction result, and the MLP model is used to output the recognition result based on the feature vector.
[0047] The online sorting method and system based on a color sorter proposed in this invention extracts features from the preprocessed image of the material to be identified, obtaining color, texture, and geometric features. By classifying and identifying the material to be identified using these color, texture, and geometric features, the accuracy of identification and the sorting quality of the material to be identified can be greatly improved. Attached Figure Description
[0048] Figure 1 This is a schematic flowchart of an online sorting method based on a color sorter proposed in this invention. Detailed Implementation
[0049] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0050] Reference Figure 1 The present invention proposes an online sorting method based on a color sorter, comprising:
[0051] Receive the material to be identified;
[0052] The received materials to be identified are then identified to obtain the identification results;
[0053] Based on the identification results, reject the unqualified materials;
[0054] The process involves identifying the received materials to be identified and obtaining the identification results, specifically including:
[0055] Acquire images of the material to be identified;
[0056] Preprocess the acquired images of the materials to be identified;
[0057] Feature extraction is performed on the preprocessed image of the material to be identified to obtain the feature data of the material to be identified; the feature data of the material to be identified includes color and texture features as well as geometric features;
[0058] The dimensionality reduction and classification of the material feature data to be identified are performed to obtain the identification results.
[0059] In this invention, feature extraction is performed on the preprocessed image of the material to be identified to obtain color, texture and geometric features. The material to be identified is then classified and identified using these color, texture and geometric features, which can greatly improve the accuracy of identification and sorting quality.
[0060] It should be understood that the methods for extracting color, texture features, and geometric features in this embodiment are all existing technologies and will not be described in detail here.
[0061] In this embodiment, the geometric features include area, aspect ratio, major axis, minor axis, and equivalent diameter. By simultaneously identifying geometric features such as area, aspect ratio, major axis, minor axis, and equivalent diameter in addition to color and texture features, the accuracy of material identification and sorting quality can be greatly improved.
[0062] In this embodiment, the feature data of the material to be identified is reduced in dimensionality and classified to obtain the identification result, specifically including:
[0063] The correlation feature selection (CFS) algorithm combined with the best first search (BF) algorithm is used to reduce the dimensionality of the material feature data to be identified, and the dimensionality reduction result is obtained.
[0064] The unsupervised machine learning technique LDA is used to classify the dimensionality reduction results to obtain the identification results.
[0065] In practical implementation, the Correlation Feature Selection (CFS) algorithm only allows subsequent classifiers to use a certain number of optimal features, which improves the algorithm's speed. Moreover, considering that the number of features used by the selected classifier to predict the true class plays an important role in the time required for data processing, the CFS algorithm is combined with the Best First Search (BF) algorithm to reduce the dimensionality of the material feature data to be identified. This allows for rapid dimensionality reduction and the selection of more useful features. In addition, the unsupervised machine learning technique LDA is used to classify the dimensionality reduction results. This method is simple, easy to implement, and computationally convenient, significantly improving the algorithm's speed.
[0066] In a further embodiment, the correlation feature selection (CFS) algorithm combined with the best-first search (BF) algorithm is used to perform dimensionality reduction on the feature data of the material to be identified, resulting in the dimensionality reduction result, specifically including:
[0067] The Correlation Feature Selection (CFS) algorithm is used to select features from the material feature data to be identified, and the selected features are used as the dataset.
[0068] Calculate the feature-category and feature-feature correlation matrices from the dataset;
[0069] The feature subset space is obtained by using the Best-First Search (BF) algorithm.
[0070] Calculate the estimated values of feature subsets in the feature subset space, and find the feature subset with the largest estimated value as the optimal feature subset;
[0071] The dimensionality of the feature subset space is reduced by using the optimal feature subset, and the dimensionality reduction result is obtained.
[0072] The process involves calculating the estimated values of feature subsets in the feature subset space and identifying the feature subset with the largest estimated value as the optimal feature subset. Specifically, it can start with an empty set or the entire set. Taking an empty set as an example, there is no feature selection at the beginning, and all possible individual features are generated. The estimated values of the features are calculated, and the feature with the largest estimated value is selected to enter the empty set. Then, the feature with the second largest estimated value is selected to enter the empty set. If the estimated values of these two features are less than the original estimated values, the feature with the second largest estimated value is removed, and then the process continues in this manner until the feature subset with the largest estimated value is found as the optimal feature subset.
[0073] In a further embodiment, the dimensionality reduction result is classified using the unsupervised machine learning technique LDA to obtain the identification result, specifically including:
[0074] The LDA algorithm of unsupervised machine learning is used to perform LDA transformation on the dimensionality reduction result to obtain the feature vector;
[0075] The feature vectors are classified using a multilayer perceptron (MLP) to obtain the classification results for each layer.
[0076] The classification results of each layer are integrated to obtain the recognition results.
[0077] In practice, the dimensionality reduction result is first transformed by the LDA algorithm, and then the feature vector is classified by the multilayer perceptron (MLP), which can improve the accuracy of the classification of the material feature data to be identified.
[0078] In this embodiment, the preprocessing of the acquired material image to be identified specifically includes:
[0079] Correct the image of the material to be identified;
[0080] Image segmentation processing algorithms are used to segment the material to be identified from the background in the image of the material to be identified.
[0081] In a further embodiment, the image segmentation processing algorithm includes the optimal thresholding method, the Otsu thresholding method, and the HSI system thresholding method.
[0082] In practice, the Otsu threshold segmentation method is used to segment the material to be identified from the background in the image of the material to be identified.
[0083] In practical implementation, due to light reflection, the binary image obtained after image segmentation may contain noise regions or holes (small black dots) in the white object area. Furthermore, due to debris in the background of the imaging system, some white dots can be found on a black background. To solve this problem, in a further embodiment, after segmenting the material to be identified from the background in the image using an image segmentation algorithm, the process further includes...
[0084] Eliminate noise.
[0085] The present invention proposes an online sorting system based on a color sorter, comprising:
[0086] The feeding device is used to receive the material to be identified;
[0087] An identification device is used to identify received materials to be identified and obtain identification results.
[0088] A rejection device is used to reject unqualified materials based on the identification results;
[0089] The identification device includes:
[0090] The image acquisition module is used to acquire images of the material to be identified.
[0091] The preprocessing module is used to preprocess the acquired images of the materials to be identified.
[0092] The feature extraction module is used to extract features from the preprocessed image of the material to be identified to obtain the feature data of the material to be identified; wherein, the feature data of the material to be identified includes color and texture features as well as geometric features;
[0093] The identification module is used to reduce the dimensionality and classify the feature data of the material to be identified to obtain the identification result.
[0094] In this invention, the feature extraction module extracts features from the preprocessed image of the material to be identified, obtaining color, texture, and geometric features. The recognition module then classifies and identifies these color, texture, and geometric features, which can greatly improve the accuracy of material identification and the sorting quality.
[0095] In this embodiment, the geometric features include area, aspect ratio, major axis, minor axis, and equivalent diameter.
[0096] To improve the accuracy of material identification and sorting quality, in this embodiment, the identification module includes:
[0097] The dimensionality reduction submodule is used to perform dimensionality reduction processing on the feature data of the material to be identified by combining the correlation feature selection (CFS) algorithm with the best first search (BF) algorithm to obtain the dimensionality reduction result.
[0098] The recognition submodule is used to classify the dimensionality reduction results using the unsupervised machine learning technique LDA to obtain the recognition results.
[0099] In a further embodiment, the recognition submodule includes an LDA model and an MLP model. The LDA model is used to output a feature vector based on the dimensionality reduction result, and the MLP model is used to output a recognition result based on the feature vector.
[0100] To improve the accuracy of feature extraction, in this embodiment, the preprocessing module includes correction and image segmentation.
[0101] In a further embodiment, the preprocessing module also includes noise removal.
[0102] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. An online sorting method based on a color sorter, characterized in that, Includes the following steps: Receive the material to be identified; The received materials to be identified are then identified to obtain the identification results; Based on the identification results, reject the unqualified materials; The process involves identifying the received materials to be identified and obtaining the identification results, specifically including: Acquire images of the material to be identified; Preprocess the acquired images of the materials to be identified; Feature extraction is performed on the preprocessed image of the material to be identified to obtain the feature data of the material to be identified; the feature data of the material to be identified includes color and texture features as well as geometric features; The correlation feature selection (CFS) algorithm combined with the best first search (BF) algorithm is used to reduce the dimensionality of the material feature data to be identified, and the dimensionality reduction result is obtained. The unsupervised machine learning technique LDA is used to classify the dimensionality reduction results to obtain the identification results.
2. The online sorting method based on a color sorter according to claim 1, characterized in that, Geometric features include area, aspect ratio, major axis, minor axis, and equivalent diameter.
3. The online sorting method based on a color sorter according to claim 1, characterized in that, The correlation feature selection algorithm (CFS) combined with the best-first search (BF) algorithm is used to reduce the dimensionality of the material feature data to be identified, resulting in the following dimensionality reduction results: The Correlation Feature Selection (CFS) algorithm is used to select features from the material feature data to be identified, and the selected features are used as the dataset. Calculate the feature-category and feature-feature correlation matrices from the dataset; The feature subset space is obtained by using the Best-First Search (BF) algorithm. Calculate the estimated values of feature subsets in the feature subset space, and find the feature subset with the largest estimated value as the optimal feature subset; The dimensionality of the feature subset space is reduced by using the optimal feature subset, and the dimensionality reduction result is obtained.
4. The online sorting method based on a color sorter according to claim 1, characterized in that, The unsupervised machine learning technique LDA is used to classify the dimensionality reduction results to obtain the recognition results, which specifically include: The LDA algorithm of unsupervised machine learning is used to perform LDA transformation on the dimensionality reduction result to obtain the feature vector; The feature vectors are classified using a multilayer perceptron (MLP) to obtain the classification results for each layer. The classification results of each layer are integrated to obtain the recognition results.
5. The online sorting method based on a color sorter according to claim 1, characterized in that, The acquired images of the material to be identified are preprocessed, specifically including: Correct the image of the material to be identified; Image segmentation processing algorithms are used to segment the material to be identified from the background in the image of the material to be identified.
6. The online sorting method based on a color sorter according to claim 5, characterized in that, Image segmentation processing algorithms include the optimal thresholding method, the Otsu thresholding method, and the HSI system thresholding method.
7. The online sorting method based on a color sorter according to claim 5, characterized in that, After segmenting the material to be identified from the background in the image using an image segmentation processing algorithm, the process also includes noise reduction.
8. An online sorting system based on a color sorter, characterized in that, include: The feeding device is used to receive the material to be identified; An identification device is used to identify received materials to be identified and obtain identification results. A rejection device is used to reject unqualified materials based on the identification results; The identification device includes: The image acquisition module is used to acquire images of the material to be identified. The preprocessing module is used to preprocess the acquired images of the materials to be identified. The feature extraction module is used to extract features from the preprocessed image of the material to be identified to obtain the feature data of the material to be identified; wherein, the feature data of the material to be identified includes color and texture features as well as geometric features; The identification module is used to reduce the dimensionality and classify the feature data of the material to be identified to obtain the identification result; The identification module includes: The dimensionality reduction submodule is used to perform dimensionality reduction processing on the feature data of the material to be identified by combining the correlation feature selection (CFS) algorithm with the best first search (BF) algorithm to obtain the dimensionality reduction result. The recognition submodule is used to classify the dimensionality reduction results using the unsupervised machine learning technique LDA to obtain the recognition results.
9. The online sorting system based on a color sorter according to claim 8, characterized in that, Geometric features include area, aspect ratio, major axis, minor axis, and equivalent diameter.
10. The online sorting system based on a color sorter according to claim 8, characterized in that, The recognition submodule includes an LDA model and an MLP model. The LDA model is used to output a feature vector based on the dimensionality reduction result; the MLP model is used to output the recognition result based on the feature vector.