Radioactive waste material classification method based on machine vision

Abstract

The invention relates to the technical field of image processing, and provides a radioactive waste material classification method based on machine vision, and the method comprises the steps: collecting radioactive waste image information through an RGB-D camera, and converting the radioactive waste image information into an HSV color space; the method comprises the following steps: extracting two heterogeneous SE-ResNet features from an HSV color space by adopting an SE-ResNet neural network; splicing and fusing the two heterogeneous SE-ResNet features to obtain first feature information, and mapping the first feature information to the same semantic space; acquiring a feature matrix X and a feature matrix Y in the same semantic space, and splicing and fusing the feature matrix X and the feature matrix Y to form second feature information; and based on the second feature information, classifying the radioactive waste by using a classifier. The method is simple, and the accuracy of material classification is effectively improved.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a radioactive waste material classification method based on machine vision. Background Art

[0002] At present, most of the radioactive waste sorting work is done manually; the second is semi-automatic sorting. At present, it is necessary to study the systematic and full-process form of automatic sorting of radioactive waste in order to achieve and improve the efficiency of automatic sorting.

[0003] The process of automatic sorting of radioactive waste involves the classification of different materials. Currently, convolutional neural networks are commonly used to extract image features and classify materials using classifiers. Large-scale material datasets are usually used for training, and the performance of the model is evaluated on the test set. In the classification process, the image processing method and the data analysis method of the neural network greatly affect the accuracy of material classification. Summary of the invention

[0004] The technical problem to be solved by the present invention is: in order to improve the accuracy of material classification, a radioactive waste material classification method based on machine vision is proposed.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is as follows: a method for classifying radioactive waste materials based on machine vision, comprising:

[0006] Step S1, using an RGB-D camera to collect radioactive waste image information, and converting the radioactive waste image information into an HSV color space;

[0007] Step S2, using the SE-ResNet neural network to extract two heterogeneous SE-ResNet features from the HSV color space;

[0008] Step S3, concatenating and fusing the two heterogeneous SE-ResNet features to obtain first feature information, and mapping the first feature information to the same semantic space;

[0009] Step S4, acquiring a feature matrix X and a feature matrix Y in the same semantic space, and concatenating and fusing the feature matrix X and the feature matrix Y to form second feature information;

[0010] Step S5: Based on the second characteristic information, a classifier is used to classify the radioactive waste.

[0011] Further, step S1 includes:

[0012] Step S11, using an RGB-D camera to collect radioactive waste images, and analyzing the brightness, saturation, and hue of the radioactive waste images;

[0013] Step S12, constructing a cone model in three-dimensional coordinates, the vertical axis of the cone model represents brightness, the cross-sectional radius of the cone model represents saturation, and the angle of the cone model on the cross section represents chromaticity, and converting the radioactive waste image into the HSV color space.

[0014] Furthermore, in step S2, the two heterogeneous SE-ResNet features are SE-ResNet50 and SE-ResNet101.

[0015] Further, step S4 includes:

[0016] Step S41, construct feature matrix X and feature matrix Y in the same semantic space, feature matrix X and feature matrix Y are two heterogeneous SE-ResNet features, construct c r ∈[c1~c l ];

[0017] in, and and Respectively represent c r middle and data, and The data used for training in two heterogeneous SE-ResNet features;

[0018] Step S42, calculate the correlation coefficient ρ between the feature matrix X and the feature matrix Y:

[0019]

[0020] Among them, ω and v are and The projection vector of ; w′ and v′ are the transpose of ω and v respectively; λ w and λ v For regular

[0021] Parameters used to control the complexity of ω and v; C XY is the covariance matrix; C XX and C YY are the autocovariance matrices of X and Y respectively:

[0022] Step S43, select the data with correlation coefficient greater than the set threshold, and calculate the mapping feature matrix:

[0023]

[0024] Step S44, calculate U and V according to the mapping feature matrix:

[0025]

[0026] Step S45: concatenate and fuse U and V to generate second feature information.

[0027] Further, in step S42, C XY , C XX and C YY The calculation method is as follows:

[0028]

[0029]

[0030] In the formula, for and The total number of pairs of relationships; c l ′ is c l The transpose of x i ,y j is the data at a position in the matrix.

[0031] Furthermore, in step S45, the calculation method for generating the second feature information DVS by splicing and fusing the feature U and the feature V is as follows:

[0032]

[0033] Furthermore, in step S45, the calculation method for generating the second feature information DVS by splicing and fusing the feature U and the feature V is as follows:

[0034]

[0035] Compared with the prior art, the present invention has the following beneficial effects: the present invention converts radioactive waste image information into the HSV color space, so that other image information besides the shape can be obtained, which is convenient for providing rich data preparation for material classification; based on the SE-ResNet neural network to extract deeper color, shape and texture features, it is convenient to more accurately identify the various properties of the material in the network model; finally, the extracted features are mapped to the semantic space and the feature information is extracted for classification. The method of the present invention simply and effectively improves the accuracy of material classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of the radioactive waste material classification method based on machine vision. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0038] In the description of the present invention, it should be noted that the terms “first” and “second” are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0039] In this embodiment, the radioactive waste material classification method based on machine vision includes:

[0040] Step S1, using an RGB-D camera to collect radioactive waste image information, and converting the radioactive waste image information into an HSV color space;

[0041] Step S2, using the SE-ResNet neural network to extract two heterogeneous SE-ResNet features from the HSV color space;

[0042] Step S3, concatenating and fusing the two heterogeneous SE-ResNet features to obtain first feature information, and mapping the first feature information to the same semantic space;

[0043] Step S4, acquiring a feature matrix X and a feature matrix Y in the same semantic space, and concatenating and fusing the feature matrix X and the feature matrix Y to form second feature information;

[0044] Step S5: Based on the second characteristic information, a classifier is used to classify the radioactive waste.

[0045] In one embodiment, step S1 includes: step S11, using an RGB-D camera to collect radioactive waste images, and analyzing the brightness (V), saturation (S), and hue (H) of the radioactive waste images; step S12, constructing a cone model in three-dimensional coordinates, the vertical axis of the cone model represents brightness, the cross-sectional radius of the cone model represents saturation, and the angle of the cone model on the cross section represents chromaticity, and converting the radioactive waste image to the HSV color space. Among them, H (hue), S (saturation), and V (brightness) are used as feature inputs. The SE-ResNet neural network enhances the network's ability to learn inter-channel dependencies through the SE (Squeeze-and-Excitation) module, thereby extracting deep features related to color features in the HSV color space.

[0046] In one embodiment, in step S2, the two heterogeneous SE-ResNet features are SE-ResNet50 and SE-ResNet101.

[0047] In one embodiment, step S4 includes: step S41, constructing a feature matrix X and a feature matrix Y in the same semantic space, wherein the feature matrix X and the feature matrix Y are two heterogeneous SE-ResNet features, and constructing c r ∈[c1~c l ];in, and and Respectively represent c r middle and data, and The data used for training in two heterogeneous SE-ResNet features.

[0048] Step S42, calculate the correlation coefficient p between the feature matrix X and the feature matrix Y:

[0049]

[0050] Among them, ω and v are and The projection vector of ; w′ and v′ are the transpose of ω and v respectively; λ w and λ v is a regularization parameter used to control the complexity of ω and v; C XY is the covariance matrix; C XX and C YY are the autocovariance matrices of X and Y respectively:

[0051] In step S42, C XY , C XX and C YY The calculation method is as follows:

[0052]

[0053] In the formula, for and The total number of pairs of relationships; c l ′ is c l The transpose of x i ,y j is the data at a position in the matrix.

[0054] The interdependent features are identified through the correlation coefficient, so that these dependent features can be selectively fused to enhance the performance of the network model. The optimal projection vectors w and v are found so that they can maximize the correlation between the feature matrix X and the feature matrix Y, while being constrained by the regularization term to avoid overfitting.

[0055] Step S43, select the data with correlation coefficient greater than the set threshold, and calculate the mapping feature matrix:

[0056]

[0057] Step S44, calculate U and V according to the mapping feature matrix:

[0058]

[0059] Step S45, concatenating and fusing U and V to generate second feature information;

[0060] In one embodiment, in step S45, the calculation method for generating the second feature information DVS by splicing and fusing the feature U and the feature V is as follows:

[0061]

[0062] In one embodiment, in step S45, the calculation method for generating the second feature information DVS by splicing and fusing the feature U and the feature V is as follows:

[0063]

[0064] In this embodiment, the mainstream deep learning network SE-ResNet neural network is used, and the SE-ResNet neural network can adaptively enhance the feature representation capability. Heterogeneous layer features are extracted based on SE-ResNet, such as SE-ResNet50 and SE-ResNet101. Based on cross-modal analysis, the Cluster-CCA model is improved, and a pre-fusion strategy is designed based on feature mapping to analyze the clustering typical correlation between heterogeneous layer features, and U and V are spliced ​​and fused to output the second feature information of deep visual semantics.

[0065] Finally, based on the second feature information, a classifier is used to classify radioactive waste; an ensemble learning method is used to fuse the results of multiple classifiers to effectively improve the material image recognition performance.

[0066] Finally, it should be noted that the above embodiments are only preferred embodiments of the present invention to illustrate the technical solutions of the present invention, rather than limiting them, and certainly not limiting the patent scope of the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention. In other words, any changes or modifications made to the main design concept and spirit of the present invention that have no substantive significance, and the technical problems they solve are still consistent with the present invention, should be included in the protection scope of the present invention. In addition, the direct or indirect application of the technical solutions of the present invention in other related technical fields is also included in the patent protection scope of the present invention.

Claims

1. A radioactive waste material classification method based on machine vision, characterized in that: include: Step S1, using an RGB-D camera to collect radioactive waste image information, and converting the radioactive waste image information into an HSV color space; Step S2, using the SE-ResNet neural network to extract two heterogeneous SE-ResNet features from the HSV color space; Step S3, concatenating and fusing the two heterogeneous SE-ResNet features to obtain first feature information, and mapping the first feature information to the same semantic space; Step S4, acquiring a feature matrix X and a feature matrix Y in the same semantic space, and concatenating and fusing the feature matrix X and the feature matrix Y to form second feature information; Step S5: Based on the second characteristic information, a classifier is used to classify the radioactive waste.

2. The method for classifying radioactive waste materials based on machine vision according to claim 1, characterized in that: Step S1 includes: Step S11, using an RGB-D camera to collect radioactive waste images, and analyzing the brightness, saturation, and hue of the radioactive waste images; Step S12, constructing a cone model in three-dimensional coordinates, the vertical axis of the cone model represents brightness, the cross-sectional radius of the cone model represents saturation, and the angle of the cone model on the cross section represents chromaticity, and converting the radioactive waste image into the HSV color space.

3. The method for classifying radioactive waste materials based on machine vision according to claim 1, characterized in that: In step S2, the two heterogeneous SE-ResNet features are SE-ResNet50 and SE-ResNet101.

4. The method for classifying radioactive waste materials based on machine vision according to claim 1, characterized in that: Step S4 includes: Step S41, construct feature matrix X and feature matrix Y in the same semantic space, feature matrix X and feature matrix Y are two heterogeneous SE-ResNet features, construct in, and and Respectively represent c r middle and data, and The data used for training in two heterogeneous SE-ResNet features; Step S42, calculate the correlation coefficient ρ between the feature matrix X and the feature matrix Y: Among them, ω and v are and The projection vector of ; w′ and v′ are the transpose of ω and v respectively; λ w and λ v is a regularization parameter used to control the complexity of ω and v; C XY is the covariance matrix; C XX and C YY are the autocovariance matrices of X and Y respectively; Step S43, select the data with correlation coefficient greater than the set threshold, and calculate the mapping feature matrix: Step S44, calculate U and V according to the mapping feature matrix: Step S45: concatenate and fuse U and V to generate second feature information.

5. The method for classifying radioactive waste materials based on machine vision according to claim 4, characterized in that: In step S42, C XY ,C XX and C YY The calculation method is as follows: In the formula, for and The total number of pairs of relationships; c l ′ is c l The transpose of x i ,y j is the data at a position in the matrix.

6. The method for classifying radioactive waste materials based on machine vision according to claim 4, characterized in that: In step S45, the calculation method for generating the second feature information DVS by splicing and fusing the feature U and the feature V is as follows:

7. The method for classifying radioactive waste materials based on machine vision according to claim 4, characterized in that: In step S45, the calculation method for generating the second feature information DVS by splicing and fusing the feature U and the feature V is as follows:

Images