Organic matter identification method and system based on pseudo-color enhancement and channel fusion

Abstract

The invention provides an organic matter identification method and system based on pseudo-color enhancement and channel fusion. The method comprises the steps that a training sample composed of marked first pseudo-color X-ray images is acquired; inputting the training sample into the improved target recognition model for training to obtain a trained organic matter detection model; wherein the improved target recognition model comprises a preprocessing module and a target detection module, and the preprocessing module comprises a channel fusion unit for separating and fusing RGB channels of a pseudo-color X-ray image and a pseudo-color enhancement unit for performing dynamic weight enhancement on the RGB channels of the pseudo-color X-ray image; the target detection module is used for identifying the preprocessed pseudo-color X-ray image; and inputting a second pseudo-color X-ray image to be identified into the organic matter detection model to obtain a target detection result of the second pseudo-color X-ray image. According to the method, the identification rate of the pseudo-color organic matter target can be remarkably improved on the premise of ensuring the real-time performance.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to an organic matter identification method and system based on pseudo-color enhancement and channel fusion. Background Art

[0002] Pseudo-color X-ray images are widely used in security inspection, industrial inspection and other fields. By assigning pseudo-color channels to object characteristics at different energy levels, the material and structure of the object can be visualized. However, due to the complex color distribution of organic matter in pseudo-color images and the low contrast with the background, the existing target detection algorithms have high false detection and missed detection rates in this field. Summary of the invention

[0003] In order to solve the above problems in the prior art, the present invention provides an organic matter recognition method and system based on pseudo-color enhancement and channel fusion, which significantly improves the recognition rate of organic matter.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is: In a first aspect, the present invention provides an organic matter identification method based on pseudo-color enhancement and channel fusion, comprising the steps of: S1, obtaining a training sample consisting of a first pseudo-color X-ray image that has been marked; S2, inputting the training samples into the improved target recognition model for training to obtain a trained organic matter detection model; The improved target recognition model includes a preprocessing module and a target detection module. The preprocessing module includes a channel fusion unit and a pseudo-color enhancement unit. The channel fusion unit separates and fuses the RGB channels of the pseudo-color X-ray image. The pseudo-color enhancement unit performs dynamic weight enhancement on the RGB channels of the pseudo-color X-ray image. The dynamic weight is obtained by cyclic optimization based on the self-attention mechanism on the basis of preset fixed weights. The target detection module is used to identify the preprocessed pseudo-color X-ray image. S3. Input the second pseudo-color X-ray image to be identified into the organic matter detection model to obtain a target detection result of the second pseudo-color X-ray image.

[0005] The beneficial effects of the present invention are as follows: by separating the RGB channels of the pseudo-color X-ray image, the influence of the RGB channels on the results can be enhanced during target recognition, and by dynamically adjusting the channel weights of the pseudo-color X-ray image to enhance the target area features, the recognition rate of pseudo-color organic targets can be significantly improved while ensuring real-time performance.

[0006] Optionally, the processing process of the channel fusion unit is as follows: S21. Reduce the dimension of the RGB channels of the pseudo-color X-ray image by principal component analysis to obtain the principal components , the formula for principal component analysis is as follows: ; S22, the main component Perform nonlinear transformation and stacking with the original RGB channels to generate a high-dimensional feature tensor of five channels and obtain a high-dimensional feature array as follows: ; S23, compressing the high-dimensional feature array through a convolutional network, mapping the high-dimensional feature tensor of five channels to a fixed three-channel input tensor : ; In the above steps S21 to S23, They are the red channel, green channel and blue channel of the pseudo-color X-ray image; is a convolution operation.

[0007] According to the above description, we can know that firstly, PCA dimensionality reduction is used to retain the main features of the pseudo-color X-ray image while reducing data redundancy; then the high-dimensional feature tensor is used to enhance the high-order feature expression ability of the pseudo-color image; finally, the convolution operation is used to reduce the number of parameters, so that the target detection module can maintain high computational efficiency while processing high-dimensional data.

[0008] Optionally, the processing process of the pseudo color enhancement unit is as follows: S24, enhancing the RGB channels of the pseudo-color X-ray image according to a pseudo-color enhancement formula, wherein the pseudo-color enhancement formula is as follows: ; in, is the pseudo-color X-ray image enhancement value, They are the red channel, green channel and blue channel of the pseudo-color X-ray image; are the weights of the first-order terms respectively; S25, dynamically optimizing the weight of the primary term through a dynamic weight allocation strategy based on a self-attention mechanism, wherein the dynamic weight The calculation formula is as follows: ; ; Among them, σ is the Sigmoid function; is a learnable parameter matrix; ReLU is the activation function; g(I) is the global feature extraction function, which extracts the global feature value for each channel of the pseudo-color X-ray image ; c is one of the red channel R, green channel G and blue channel B; is the value of pixel (i, j) in channel c; S26, obtain the image after dynamic weight enhancement for: .

[0009] According to the above description, the adaptive enhancement of pseudo-color X-ray images is achieved through the dynamic weight allocation strategy of the self-attention mechanism.

[0010] Optionally, when the improved target recognition model starts training, the first-order term weight corresponding to each recognition category is a preset fixed weight.

[0011] According to the above description, by presetting fixed weights, the target characteristics of the image can be quickly enhanced, and the background and noise can be preliminarily screened, thereby improving the training efficiency.

[0012] Optionally, when the identification category is organic matter, the preset fixed weights of its red channel and green channel are greater than the preset fixed weight of the blue channel; when the identification category is inorganic matter, the preset fixed weight of its red channel is less than the preset fixed weights of the red channel and the blue channel.

[0013] Optionally, the identification categories include plastics, liquids, foods, inorganic substances and noise areas.

[0014] Optionally, the loss function of the improved target recognition model during training also includes adjusting classification loss weights according to target detection difficulty and recognition category distribution.

[0015] According to the above description, by adding a dynamic category weighting strategy, the weight balance between categories can be dynamically optimized to enhance the classification and positioning capabilities of complex targets.

[0016] Optionally, the step S1 includes: An original sample consisting of a marked first pseudo-color X-ray image is obtained, the original sample is expanded by a multi-view enhancement technology to obtain an expanded sample, and the original sample and the training sample are combined to obtain a training sample.

[0017] Optionally, the expanding the original sample by using the multi-view enhancement technology further includes: S11, converting the first pseudo color image from the RGB space to the HSI space, and adding random perturbations to the hue, saturation and brightness; S12, randomly performing geometric transformations including rotation, scaling and cropping on the first pseudo-color image; S13, adding Poisson noise and Gaussian blur to the first pseudo-color image.

[0018] According to the above description, multi-view enhancement technology, including color perturbation, geometric transformation and noise simulation, is used to expand the diversity of data samples and improve the generalization ability of the target detection model.

[0019] In a second aspect, an organic matter identification system based on pseudo-color enhancement and channel fusion includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the organic matter identification method based on pseudo-color enhancement and channel fusion of the first aspect when executing the computer program.

[0020] Among them, the technical effect corresponding to the organic matter identification system based on pseudo color enhancement and channel fusion provided by the second aspect refers to the relevant description of the organic matter identification method based on pseudo color enhancement and channel fusion provided by the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart of an organic matter identification method based on pseudo color enhancement and channel fusion according to an embodiment of the present invention; Figure 2 It is a schematic diagram of the framework of an organic matter identification system based on pseudo-color enhancement and channel fusion according to an embodiment of the present invention; Description of reference numerals: 1. Organic matter identification system based on pseudo-color enhancement and channel fusion; 2. Processor; 3. Memory. DETAILED DESCRIPTION

[0022] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0023] Embodiment 1 This embodiment is suitable for application scenarios such as logistics, warehousing and express delivery that require X-ray inspection of packages. In particular, for packages subjected to three-level X-ray irradiation, the existing target detection algorithms have high false detection and missed detection rates in this field. This embodiment separates the RGB channels of the pseudo-color X-ray image and combines the self-attention mechanism to adjust the pseudo-color weight ratio in real time and perform refined enhancement on the target area, thereby significantly improving the recognition rate of pseudo-color organic targets while ensuring real-time performance. See the following description for details.

[0024] It should be noted that the three-level X-ray irradiation of the package means that three beams of X-ray are irradiated on the package respectively, and different colors are displayed according to the penetration degree of different objects in the package. Among them, the RGB channels of the pseudo-color image correspond to different energy level information of the object.

[0025] Please refer to Figure 1 , an organic matter identification method based on pseudo color enhancement and channel fusion, comprising the steps of: S1. Obtain a training sample consisting of a first labeled pseudo-color X-ray image.

[0026] In this embodiment, step S1 includes: An original sample consisting of a marked first pseudo-color X-ray image is obtained, the original sample is expanded by a multi-view enhancement technology to obtain an expanded sample, and the original sample and the training sample are combined to obtain a training sample.

[0027] Among them, the expansion of the original sample through multi-view enhancement technology also includes: S11. Convert the first pseudo color image from the RGB space to the HSI space, and add random perturbations to the hue, saturation, and brightness.

[0028] Among them, the ‌HSI space‌ is a color space based on the human visual system, which uses hue, saturation, and intensity to describe colors.

[0029] Among them, adding random perturbations to hue, saturation, and brightness means adding random perturbation values ​​to the original values. For example, saturation S is added with random perturbation value δ S Then we get the saturation after random disturbance S'=S+δ S .

[0030] Therefore, different imaging conditions are simulated by random perturbations to expand the diversity of data samples.

[0031] S12, randomly applying geometric transformations including rotation, scaling and cropping to the first pseudo-color image.

[0032] Among them, by applying geometric transformations such as random rotation, scaling and cropping to the image, different imaging conditions are simulated to further expand the diversity of data samples.

[0033] S13, adding Poisson noise and Gaussian blur to the first pseudo-color image.

[0034] Among them, Poisson noise and Gaussian blur are added to enhance the robustness of the model to real noise.

[0035] Therefore, multi-view enhancement technology is used to expand the diversity of data samples, improve the model's adaptability to complex scenes and different materials, and significantly enhance the generalization ability.

[0036] S2. Input the training samples into the improved target recognition model for training to obtain a trained organic matter detection model.

[0037] Among them, the improved target recognition model includes a preprocessing module and a target detection module. The preprocessing module includes a channel fusion unit and a pseudo-color enhancement unit. The channel fusion unit separates and fuses the RGB channels of the pseudo-color X-ray image. The pseudo-color enhancement unit dynamically weights the RGB channels of the pseudo-color X-ray image. The dynamic weights are obtained by cyclic optimization based on the self-attention mechanism on the basis of preset fixed weights. The target detection module is used to identify the preprocessed pseudo-color X-ray image.

[0038] In this embodiment, the target detection module uses the YOLOv5 model. The prior art usually directly uses standard RGB images as input, so there is a problem of low organic recognition rate in the processing of pseudo-color X-ray images. This embodiment optimizes the data quality of the pseudo-color image by performing RGB channel separation and fusion and channel dynamic weight allocation on the input pseudo-color X-ray image, providing a more accurate input for the target detection model.

[0039] Therefore, in this embodiment, the processing process of the channel fusion unit is as follows: S21. Reduce the dimension of the RGB channels of the pseudo-color X-ray image by principal component analysis to obtain the principal components , the formula for principal component analysis is as follows: .

[0040] Therefore, step S21 retains the main features of the pseudo-color X-ray image through PCA dimensionality reduction while reducing data redundancy.

[0041] S22, the main component Perform nonlinear transformation and stacking with the original RGB channels to generate a high-dimensional feature tensor of five channels and obtain a high-dimensional feature array as follows: .

[0042] Therefore, step S22 enhances the high-order feature expression capability of the pseudo-color image through the high-dimensional feature tensor.

[0043] S23. Compress the high-dimensional feature array through the convolutional network and map the high-dimensional feature tensor of five channels to a fixed three-channel input tensor. : ; In the above steps S21 to S23, They are the red channel, green channel and blue channel of the pseudo-color X-ray image; is a convolution operation.

[0044] Therefore, step S23 reduces the number of parameters through convolution operation, so that the target detection module maintains high computational efficiency while processing high-dimensional data.

[0045] In this embodiment, the processing process of the pseudo color enhancement unit is as follows: S24, enhancing the RGB channels of the pseudo-color X-ray image according to a pseudo-color enhancement formula, wherein the pseudo-color enhancement formula is as follows: ; in, is the pseudo-color X-ray image enhancement value, They are the red channel, green channel and blue channel of the pseudo-color X-ray image; are the weights of the first-order terms respectively.

[0046] S25, dynamically optimizing the weight of the primary term through a dynamic weight allocation strategy based on a self-attention mechanism, wherein the dynamic weight The calculation formula is as follows: ; ; Among them, σ is the Sigmoid function; is a learnable parameter matrix; ReLU is the activation function; g(I) is the global feature extraction function, which extracts the global feature value for each channel of the pseudo-color X-ray image ; c is one of the red channel R, green channel G and blue channel B; is the value of pixel (i, j) in channel c.

[0047] Therefore, the limitations of fixed weights are avoided through the dynamic weight allocation strategy of the self-attention mechanism, and adaptive enhancement of pseudo-color X-ray images is achieved.

[0048] S26, obtain the image after dynamic weight enhancement for: .

[0049] Therefore, the image I after dynamic weight enhancement final It can adaptively enhance the key feature areas of pseudo-color X-ray images.

[0050] In this embodiment, when the improved target recognition model starts training, the first-order weight corresponding to each recognition category is a preset fixed weight. The recognition categories include plastic, liquid, food, inorganic matter and noise area, as shown in Table 1.

[0051] Table 1. Preset fixed weights for different recognition categories

[0052] As shown in Table 1 above, when the identification category is organic matter, the preset fixed weights of its red channel and green channel are greater than the preset fixed weight of the blue channel. When the identification category is inorganic matter, the preset fixed weight of its red channel is less than the preset fixed weights of the red channel and the blue channel. Specifically, plastics are mostly shown as bright red areas in pseudo-color X-ray images, so the red channel has the highest weight. Liquids are usually shown as mixed red and green areas in pseudo-color X-ray images, so the weights of the red channel and the green channel are high. Food is usually concentrated in the red channel in pseudo-color X-ray images, so the red channel has the highest weight. Inorganic matter is mostly distributed in green or blue, so the weights of the blue channel and the green channel are high. As for the balanced distribution of background noise in the noise area, the weights of the three channels are relatively low. Therefore, the image target characteristics are quickly enhanced by presetting fixed weights, and the background and noise are preliminarily screened, thereby improving the training efficiency.

[0053] In this embodiment, the loss function of the improved target recognition model during training also includes adjusting the classification loss weight according to the target detection difficulty and the recognition category distribution to dynamically optimize the weight balance between categories and enhance the classification and positioning capabilities of complex targets.

[0054] Thus, the pseudo-color enhanced X-ray image is input into the YOLOv5 model, the dynamic weight of the recognition category is optimized by the loss function, and iterative training is performed according to the set learning rate and number of iterations, and finally a trained organic detection model is obtained. At this time, the performance of the organic detection model of this embodiment is compared with the performance of the existing organic recognition model, and the following Table 2 is obtained.

[0055] Table 2. Recognition performance of different recognition categories before and after optimization

[0056] Therefore, compared with the existing organic matter recognition model, the organic matter detection model of this embodiment has significantly improved the model accuracy, recall rate, and mAP (Mean Average Precision) by about 5%.

[0057] S3. Input the second pseudo-color X-ray image to be identified into an organic matter detection model to obtain a target detection result of the second pseudo-color X-ray image.

[0058] In summary, this embodiment is comprehensively optimized for the pseudo-color organic matter detection scenario, so that the recognition rate of pseudo-color organic matter targets can be significantly improved while ensuring real-time performance, thereby enhancing the practical application value.

[0059] Embodiment 2 Please refer to Figure 2 The organic matter identification system 1 based on pseudo-color enhancement and channel fusion includes a memory 3, a processor 2, and a computer program stored in the memory 3 and executable on the processor 2. When the processor 2 executes the computer program, the steps in the above-mentioned embodiment 1 are implemented.

[0060] Since the system / device described in the above embodiments of the present invention is a system / device used to implement the method of the above embodiments of the present invention, a person skilled in the art can understand the specific structure and deformation of the system / device based on the method described in the above embodiments of the present invention, and thus will not be described in detail here. All systems / devices used in the method of the above embodiments of the present invention belong to the scope of protection of the present invention.

[0061] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, apparatuses or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0062] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (apparatus) and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions.

[0063] It should be noted that in the claims, any reference numerals placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In the claims enumerating several means, several of these means may be embodied by the same hardware. The use of the words first, second, third, etc., is for convenience of expression only and does not indicate any order. These words may be understood as part of the component name.

[0064] In addition, it should be noted that, in the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0065] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0066] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include these modifications and variations.

Claims

1. An organic matter identification method based on pseudo-color enhancement and channel fusion, characterized in that: Includes steps: S1, obtaining a training sample consisting of a first pseudo-color X-ray image that has been marked; S2, inputting the training samples into the improved target recognition model for training to obtain a trained organic matter detection model; The improved target recognition model includes a preprocessing module and a target detection module. The preprocessing module includes a channel fusion unit and a pseudo-color enhancement unit. The channel fusion unit separates and fuses the RGB channels of the pseudo-color X-ray image. The pseudo-color enhancement unit performs dynamic weight enhancement on the RGB channels of the pseudo-color X-ray image. The dynamic weight is obtained by cyclic optimization based on the self-attention mechanism on the basis of preset fixed weights. The target detection module is used to identify the preprocessed pseudo-color X-ray image. S3. Input the second pseudo-color X-ray image to be identified into the organic matter detection model to obtain a target detection result of the second pseudo-color X-ray image.

2. The method for identifying organic matter based on pseudo-color enhancement and channel fusion according to claim 1, characterized in that: The processing process of the channel fusion unit is as follows: S21. Reduce the dimension of the RGB channels of the pseudo-color X-ray image by principal component analysis to obtain the principal components , the formula for principal component analysis is as follows: ； S22, the main component Perform nonlinear transformation and stacking with the original RGB channels to generate a high-dimensional feature tensor of five channels and obtain a high-dimensional feature array as follows: ； S23, compressing the high-dimensional feature array through a convolutional network, mapping the high-dimensional feature tensor of five channels to a fixed three-channel input tensor : ； In the above steps S21 to S23, They are the red channel, green channel and blue channel of the pseudo-color X-ray image; is a convolution operation.

3. The method for identifying organic matter based on pseudo-color enhancement and channel fusion according to claim 1, characterized in that: The processing process of the pseudo color enhancement unit is as follows: S24, enhancing the RGB channels of the pseudo-color X-ray image according to a pseudo-color enhancement formula, wherein the pseudo-color enhancement formula is as follows: ； in, is the pseudo-color X-ray image enhancement value, They are the red channel, green channel and blue channel of the pseudo-color X-ray image; are the weights of the first-order terms respectively; S25, dynamically optimizing the weight of the primary term through a dynamic weight allocation strategy based on a self-attention mechanism, wherein the dynamic weight The calculation formula is as follows: ； ； Among them, σ is the Sigmoid function; is a learnable parameter matrix; ReLU is the activation function; g(I) is the global feature extraction function, which extracts the global feature value for each channel of the pseudo-color X-ray image ; c is one of the red channel R, green channel G and blue channel B; is the value of pixel (i, j) in channel c; S26, obtain the image after dynamic weight enhancement for: 。 4. The method for identifying organic matter based on pseudo-color enhancement and channel fusion according to claim 3, characterized in that: When the improved target recognition model starts training, the first-order item weight corresponding to each recognition category is a preset fixed weight.

5. The method for identifying organic matter based on pseudo-color enhancement and channel fusion according to claim 4, characterized in that: When the identification category is organic matter, the preset fixed weights of its red channel and green channel are greater than the preset fixed weight of the blue channel. When the identification category is inorganic matter, the preset fixed weight of its red channel is less than the preset fixed weights of the red channel and the blue channel.

6. The method for identifying organic matter based on pseudo-color enhancement and channel fusion according to claim 4, characterized in that: The identification categories include plastics, liquids, food, inorganics, and noise areas.

7. The method for identifying organic matter based on pseudo color enhancement and channel fusion according to any one of claims 1 to 6, characterized in that: The loss function of the improved target recognition model during training also includes adjusting the classification loss weight according to the target detection difficulty and the recognition category distribution.

8. The method for identifying organic matter based on pseudo-color enhancement and channel fusion according to any one of claims 1 to 6, characterized in that: The step S1 comprises: An original sample consisting of a marked first pseudo-color X-ray image is obtained, the original sample is expanded by a multi-view enhancement technology to obtain an expanded sample, and the original sample and the training sample are combined to obtain a training sample.

9. The method for identifying organic matter based on pseudo-color enhancement and channel fusion according to claim 8, characterized in that: The expanding the original sample by using the multi-view enhancement technology also includes: S11, converting the first pseudo color image from the RGB space to the HSI space, and adding random perturbations to the hue, saturation and brightness; S12, randomly performing geometric transformations including rotation, scaling and cropping on the first pseudo-color image; S13, adding Poisson noise and Gaussian blur to the first pseudo-color image.

10. An organic matter identification system based on pseudo-color enhancement and channel fusion, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for identifying organic matter based on pseudo-color enhancement and channel fusion according to any one of claims 1 to 9 is implemented.

Images