Multi-dimensional cooperative adaptive image edge detection method, device, equipment and medium

By improving the VRWKV model into an edge prediction model and employing multi-dimensional information fusion and dual-channel entropy change feedforward technology, the efficiency and robustness issues of the ViT and VRWKV models in high-resolution image detection were resolved, achieving efficient and accurate edge detection.

CN122657178BActive Publication Date: 2026-11-10HANGZHOU ANMAISHENG INTELLIGENT TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611125077.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-07-28
Publication Date
2026-11-10
Estimated Expiration
2046-07-28

AI Technical Summary

Technical Problem

Existing ViT models suffer from high computational complexity and low efficiency in high-resolution image edge detection. VRWKV models lack robustness and detection accuracy when facing weak edges and complex texture interference in industrial scenarios, making it difficult to simultaneously achieve efficiency, accuracy, and robustness.

Method used

The VRWKV model is improved into an edge prediction model by introducing a multi-dimensional information fusion module and a dual-channel entropy change feedforward module. By processing value transfer, gradient and context feature maps in parallel, a spatial preprocessed map is generated, and feature compression and fidelity map fusion are performed. Finally, the decoder performs upsampling to generate an edge prediction image.

Benefits of technology

It significantly improves the efficiency, accuracy, and robustness of image edge detection, enabling accurate detection of weak edges and textures in complex industrial environments, meeting the high reliability and high precision online detection requirements of the nuclear power industry.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122657178B_ABST
    Figure CN122657178B_ABST
Patent Text Reader

Abstract

Embodiments of the present application disclose a multi-dimensional cooperative adaptive image edge detection method, device, equipment and medium, comprising: acquiring an edge prediction model comprising an encoder and a decoder; wherein the encoder comprises a multi-dimensional information fusion module and a double-channel entropy variation feedforward module; determining a value transmission feature map, a gradient feature map and a context feature map corresponding to the image to be detected in parallel through the multi-dimensional information fusion module, and generating a spatial preprocessing map corresponding to the image to be detected based on the value transmission feature map, the gradient feature map and the context feature map; determining a feature compression map and a feature fidelity map corresponding to the spatial preprocessing map in parallel through the double-channel entropy variation feedforward module, and determining an encoded feature map corresponding to the spatial preprocessing map based on the feature compression map and the feature fidelity map; and performing up-sampling processing on the encoded feature map through the decoder to obtain an edge prediction image corresponding to the image to be detected, which significantly improves the efficiency, accuracy and robustness of image edge detection.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and in particular to a multidimensional collaborative adaptive image edge detection method, apparatus, device, and medium. Background Technology

[0002] In the nuclear power industry, the edge and weld integrity of critical equipment such as pressure vessels and fuel rods are core technological indicators that determine their structural reliability, service life, and nuclear safety. However, during the precision manufacturing and long-term service of these critical equipment, industrial imaging is limited by factors such as acquisition rate, depth of field, and production line vibration, which can easily lead to problems such as blurred image edges, insufficient contrast, and difficulty in identifying minute defects. These issues fundamentally restrict the detection accuracy of subsequent automated inspection algorithms.

[0003] In existing technologies, the Vision Transformer (ViT) model or the Vision Receptance Weighted Key Value (VRWKV) model is often used to perform edge detection on industrial images of key components.

[0004] However, due to the quadratic computational complexity of the core self-attention mechanism of the ViT model, it incurs huge computational and memory overhead when processing high-resolution images from industrial cameras, making it difficult to meet the stringent requirements of industrial production lines for edge detection speed. Secondly, although the VRWKV model achieves linear computational efficiency proportional to the input image size while maintaining the global receptive field, its architecture is still biased towards general feature extraction. Therefore, its robustness and detection accuracy in the face of weak edges and complex texture interference commonly found in industrial scenarios still need improvement. Summary of the Invention

[0005] This invention provides a multidimensional collaborative adaptive image edge detection method, apparatus, device, and medium, which solves the problem that it is difficult to simultaneously achieve efficiency, accuracy, and robustness when using the ViT model or VRWKV model for image edge detection, and significantly improves the efficiency, accuracy, and robustness of image edge detection.

[0006] In a first aspect, embodiments of the present invention provide a multidimensional collaborative adaptive image edge detection method, comprising: acquiring an image to be detected and a pre-constructed edge prediction model; wherein the edge prediction model includes an encoder and a decoder, and the encoder includes a multidimensional information fusion module and a dual-channel entropy-change feedforward module; through the multidimensional information fusion module, a value transfer feature map, a gradient feature map, and a context feature map corresponding to the image to be detected are determined in parallel, and a spatial preprocessing map corresponding to the image to be detected is generated based on the value transfer feature map, the gradient feature map, and the context feature map; through the dual-channel entropy-change feedforward module, a feature compression map and a feature fidelity map corresponding to the spatial preprocessing map are determined in parallel, and feature fusion is performed on the feature compression map and the feature fidelity map to obtain an encoded feature map corresponding to the spatial preprocessing map; through the decoder, the encoded feature map is upsampled to obtain an edge prediction image corresponding to the image to be detected.

[0007] Secondly, embodiments of the present invention also provide a multi-dimensional collaborative adaptive image edge detection device, comprising: an information acquisition module for acquiring an image to be detected and a pre-constructed edge prediction model; wherein the edge prediction model includes an encoder and a decoder, the encoder including a multi-dimensional information fusion module and a dual-channel entropy-change feedforward module; a spatial preprocessing module for determining, in parallel, a value transfer feature map, a gradient feature map, and a context feature map corresponding to the image to be detected through the multi-dimensional information fusion module, and generating a spatial preprocessed map corresponding to the image to be detected based on the value transfer feature map, gradient feature map, and context feature map; an image encoding module for determining, in parallel, a feature compression map and a feature fidelity map corresponding to the spatial preprocessed map through the dual-channel entropy-change feedforward module, and performing feature fusion on the feature compression map and the feature fidelity map to obtain an encoded feature map corresponding to the spatial preprocessed map; and an image decoding module for upsampling the encoded feature map through the decoder to obtain an edge prediction image corresponding to the image to be detected.

[0008] Thirdly, embodiments of the present invention also provide an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to execute the multidimensional collaborative adaptive image edge detection method provided in any embodiment of the present invention.

[0009] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the multidimensional collaborative adaptive image edge detection method of any embodiment of the present invention.

[0010] The technical solution of this invention acquires an image to be detected and a pre-constructed edge prediction model. The edge prediction model includes an encoder and a decoder. The encoder includes a multi-dimensional information fusion module and a dual-channel entropy-change feedforward module. The multi-dimensional information fusion module determines, in parallel, the value transfer feature map, gradient feature map, and context feature map corresponding to the image to be detected, and generates a spatial preprocessing map corresponding to the image to be detected based on these maps. The dual-channel entropy-change feedforward module determines, in parallel, the feature compression map and feature fidelity map corresponding to the spatial preprocessing map, and performs feature fusion on these maps to obtain an encoded feature map corresponding to the spatial preprocessing map. The decoder upsamples the encoded feature map to obtain the edge prediction image corresponding to the image to be detected. This method solves the problem that using the ViT model or VRWKV model for image edge detection is difficult to simultaneously achieve efficiency, accuracy, and robustness, significantly improving the efficiency, accuracy, and robustness of image edge detection.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a multi-dimensional collaborative adaptive image edge detection method provided in Embodiment 1 of the present invention.

[0014] Figure 2 This is a schematic diagram of an edge prediction model architecture provided by an embodiment of the present invention.

[0015] Figure 3 This is a flowchart of another multidimensional collaborative adaptive image edge detection method provided in Embodiment 2 of the present invention.

[0016] Figure 4 This is a schematic diagram illustrating the execution flow of a multi-dimensional information fusion module according to an embodiment of the present invention.

[0017] Figure 5 This is a schematic diagram illustrating a value transfer process according to an embodiment of the present invention.

[0018] Figure 6 This is a schematic diagram illustrating the execution flow of a dual-channel entropy change feedforward module according to an embodiment of the present invention.

[0019] Figure 7 This is a schematic diagram of the structure of a multi-dimensional collaborative adaptive image edge detection device according to Embodiment 3 of the present invention.

[0020] Figure 8 This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] Example 1 Figure 1 This is a flowchart of a multidimensional collaborative adaptive image edge detection method according to Embodiment 1 of the present invention. This embodiment is applicable to edge detection of images. The method can be executed by a multidimensional collaborative adaptive image edge detection device, which can be implemented in hardware and / or software and can be configured in an electronic device such as a computer.

[0024] like Figure 1 As shown in the figure, the multidimensional collaborative adaptive image edge detection method disclosed in this embodiment includes S110-S140.

[0025] S110. Acquire the image to be detected and the pre-built edge prediction model; wherein the edge prediction model includes an encoder and a decoder, and the encoder includes a multi-dimensional information fusion module and a dual-channel entropy change feedforward module.

[0026] In this embodiment, the image to be detected can be understood as an image that requires edge detection, which is often a measured industrial image of key equipment in the nuclear power field. The edge prediction model can be understood as a model obtained by improving the VRWKV model.

[0027] In this step, specifically, to improve the robustness and accuracy of edge detection, the VRWKV model, which only includes an encoder and contains only a spatial information interaction module and a channel feature exchange module, can be improved. Specifically, the spatial information interaction module in the VRWKV model can be replaced with a multi-dimensional information fusion module, the channel feature exchange module can be replaced with a dual-channel entropy-change feedforward module, and a decoder containing an upsampling module can be added to the VRWKV model, thereby obtaining... Figure 2 The edge prediction model shown.

[0028] The advantages of this setup are twofold: First, compared to existing techniques using the ViT model for image edge detection, the solution in this embodiment, by improving the VRWKV model to obtain an edge prediction model, can further optimize model performance while inheriting the original efficiency advantages of the VRWKV model, thereby achieving a dual improvement in efficiency and accuracy. Second, by replacing the spatial information interaction module in the VRWKV model with a multi-dimensional information fusion module, the single spatial information preprocessing operation in the VRWKV model can be replaced, providing a richer and more powerful input for subsequent global information mixing. Finally, by replacing the channel feature exchange module in the VRWKV model with a dual-channel entropy-change feedforward module, the fixed and single feedforward network in the VRWKV model can be replaced, thereby significantly improving the final detection accuracy of the model in complex industrial environments.

[0029] S120. Through the multi-dimensional information fusion module, the value transfer feature map, gradient feature map and context feature map corresponding to the image to be detected are determined in parallel, and a spatial preprocessing map corresponding to the image to be detected is generated based on the value transfer feature map, gradient feature map and context feature map.

[0030] In this embodiment, the value-transfer feature map can be understood as a feature map formed by injecting the values ​​of the directly adjacent neighboring pixels of each pixel in the image to be detected into itself. The gradient feature map can be understood as a feature map formed by performing a gradient extraction operation based on the center difference on the image to be detected. The context feature map can be understood as a feature map formed by performing a context extraction operation on the image to be detected.

[0031] In this step, specifically, value transfer processing, gradient extraction based on center difference, and context extraction can be performed simultaneously on the image to be detected, resulting in a value transfer feature map, gradient feature map, and context feature map corresponding to the image to be detected. Then, the value transfer feature map, gradient feature map, and context feature map can be integrated to obtain a spatial preprocessed map corresponding to the image to be detected.

[0032] S130. Through the dual-channel entropy change feedforward module, the feature compression map and feature fidelity map corresponding to the spatial preprocessing map are determined in parallel, and the feature compression map and feature fidelity map are fused to obtain the encoded feature map corresponding to the spatial preprocessing map.

[0033] In this embodiment, the feature compression map can be understood as a map obtained by refining and reconstructing the spatial preprocessed map. The feature fidelity map can be understood as a map obtained by performing a nonlinear transformation on the spatial preprocessed map.

[0034] In this step, specifically, a compressed-size feedforward network and a standard-size feedforward network can be obtained in the dual-channel entropy-change feedforward module. Then, the compressed-size feedforward network can be used to determine the feature compression map corresponding to the spatial preprocessed map, and the standard-size feedforward network can be used to determine the feature fidelity map corresponding to the spatial preprocessed map.

[0035] The advantage of this setup is that, since the VRWKV model's architecture is still biased towards general feature extraction, its internal spatial information interaction module and channel feature transformation module lack specialized perception mechanisms for structured information such as edges, gradients, and textures. As a result, its robustness and detection accuracy still need to be improved when facing weak edge and complex texture interference common in industrial scenarios. Therefore, the technical solution in this embodiment determines the value transfer feature map, gradient feature map, and context feature map corresponding to the image to be detected in parallel, and generates a spatial preprocessing map corresponding to the image to be detected based on the value transfer feature map, gradient feature map, and context feature map. This can ensure the robustness and accuracy of image edge detection when facing weak edge and complex texture interference common in industrial scenarios.

[0036] S140. The encoded feature map is upsampled using a decoder to obtain an edge prediction image corresponding to the image to be detected.

[0037] In this embodiment, the edge prediction image can be understood as an image with edge coordinates.

[0038] In this step, specifically, the encoded feature map can be upsampled using the upsampling module in the decoder to obtain an edge prediction image corresponding to the image to be detected. Alternatively, the encoded feature map can be upsampled using the upsampling module in the decoder to obtain the edge coordinates corresponding to the image to be detected. Then, based on the image to be detected and its corresponding edge coordinates, an edge prediction image is generated.

[0039] The technical solution of this embodiment acquires the image to be detected and a pre-constructed edge prediction model. The edge prediction model includes an encoder and a decoder. The encoder includes a multi-dimensional information fusion module and a dual-channel entropy-change feedforward module. The multi-dimensional information fusion module determines, in parallel, the value transfer feature map, gradient feature map, and context feature map corresponding to the image to be detected, and generates a spatial preprocessing map corresponding to the image to be detected based on these maps. The dual-channel entropy-change feedforward module determines, in parallel, the feature compression map and feature fidelity map corresponding to the spatial preprocessing map, and performs feature fusion on these maps to obtain an encoded feature map corresponding to the spatial preprocessing map. The decoder upsamples the encoded feature map to obtain the edge prediction image corresponding to the image to be detected. This approach solves the problem that using the ViT model or VRWKV model for image edge detection is difficult to simultaneously achieve high efficiency, accuracy, and robustness, significantly improving the efficiency, accuracy, and robustness of image edge detection.

[0040] Example 2 Figure 3 This is a flowchart of another multi-dimensional collaborative adaptive image edge detection method provided by Embodiment 2 of the present invention. This embodiment is a further optimization and extension based on the above embodiments, and can be combined with various optional technical solutions in the above embodiments.

[0041] like Figure 3 As shown in the figure, this embodiment discloses a multidimensional collaborative adaptive image edge detection method, including steps S210-S2100.

[0042] S210. Acquire the image to be detected and a pre-built edge prediction model; wherein the edge prediction model includes an encoder and a decoder, the encoder includes a multi-dimensional information fusion module and a dual-channel entropy change feedforward module, the multi-dimensional information fusion module includes a value transfer path, a gradient extraction path and a context-aware path, and the dual-channel entropy change feedforward module includes a feature compression path and a feature fidelity path.

[0043] Optionally, before obtaining the pre-built edge prediction model, the method further includes: acquiring training images and an edge prediction model to be trained; wherein the edge prediction model to be trained includes an encoder to be trained and a decoder to be trained, and the encoder to be trained includes a multi-dimensional information fusion module to be trained and a dual-channel entropy-change feedforward module to be trained; generating a spatial preprocessing map corresponding to the training image through the multi-dimensional information fusion module to be trained; performing feature compression or feature fidelity operation on the spatial preprocessing map through the dual-channel entropy-change feedforward module to be trained, and determining the feature encoding representation map corresponding to the training image based on the feature compression or feature fidelity result; upsampling the feature encoding representation map through the decoder to be trained to obtain an edge prediction image corresponding to the training image; determining whether the edge detection accuracy of the edge prediction model to be trained meets the preset accuracy requirements based on each training image and the edge prediction image corresponding to each training image; if not, continuing to train the edge prediction model to be trained until a trained edge prediction model is obtained.

[0044] Specifically, the dual-channel entropy-change feedforward module to be trained can randomly activate either a feature compression path or a feature fidelity path with equal probability (i.e., 50%) during each forward propagation, while setting the computational output of unselected paths to zero and excluding them from gradient backpropagation. Then, in each forward propagation, the selected feature compression path or feature fidelity path can be used to process the spatial preprocessing map to obtain the feature compression result or the feature fidelity result.

[0045] The advantage of this setup is that by optimizing the model through a random path selection mechanism, the feature compression path and the feature fidelity path can be forced to learn complementary and low-correlation feature representations. This forces the model to alternately receive two gradient signals, "lossy compression" and "lossless representation," during training, thereby significantly improving the model's generalization ability and noise resistance.

[0046] S220. Through the value transfer path, obtain each multi-channel pixel in the image to be detected, as well as the neighboring pixel group corresponding to each multi-channel pixel, and transfer the neighboring pixel value in each neighboring pixel group to the corresponding multi-channel pixel to obtain the value transfer feature map corresponding to the image to be detected.

[0047] In this step, specifically, such as Figure 4 As shown, multiple channel groups corresponding to each multi-channel pixel group and their corresponding neighboring pixel groups can be obtained through the value transfer path. Then, the neighboring pixel values ​​within each neighboring pixel group are transferred to the corresponding channel group to obtain the value transfer feature map corresponding to the image to be detected.

[0048] Optionally, through the value transfer path, each multi-channel pixel in the image to be detected, and the corresponding neighboring pixel groups, are obtained. The neighboring pixel values ​​within each neighboring pixel group are then transferred to the corresponding multi-channel pixel to obtain a value transfer feature map corresponding to the image to be detected. This includes: obtaining each multi-channel pixel in the image to be detected through the value transfer path, and determining, based on the number of feature channels corresponding to the image to be detected, the upward positive traffic channel group, downward positive traffic channel group, left positive traffic channel group, and right positive traffic channel group corresponding to each multi-channel pixel; based on the number of feature channels corresponding to each multi-channel pixel... The corresponding upward positive traffic channel group, downward positive traffic channel group, left positive traffic channel group, and right positive traffic channel group are used to determine the upward neighbor pixel group, downward neighbor pixel group, left neighbor pixel group, and right neighbor pixel group corresponding to each multi-channel pixel. The pixel values ​​in the upward positive traffic channel group, downward positive traffic channel group, left positive traffic channel group, and right positive traffic channel group corresponding to each multi-channel pixel are replaced by the corresponding upward neighbor pixel group, downward positive traffic channel group, left positive traffic channel group, and right positive traffic channel group to obtain the value transfer feature map corresponding to the image to be detected.

[0049] Specifically, the number of feature channels corresponding to the image to be detected can be used as the number of feature channels corresponding to each multi-channel pixel. Then, the number of feature channels corresponding to each multi-channel pixel can be averaged to obtain the upward positive traffic channel group, downward positive traffic channel group, left positive traffic channel group, and right positive traffic channel group corresponding to each multi-channel pixel. Each positive traffic channel group includes at least one feature channel.

[0050] Next, the top, bottom, left, and right neighbor pixel groups corresponding to each multi-channel pixel can be determined. Finally, it can be determined whether the top, bottom, left, and right neighbor pixel groups corresponding to each multi-channel pixel are empty. When none of the top, bottom, left, and right neighbor pixel groups corresponding to a multi-channel pixel are empty, the pixel values ​​in the corresponding forward channel group are replaced with the corresponding neighbor pixel groups. When any neighbor pixel group corresponding to a multi-channel pixel is empty, the pixel values ​​in the corresponding forward channel group are replaced with the non-empty neighbor pixel groups corresponding to that multi-channel pixel group, while keeping the pixel values ​​in the forward channel group corresponding to the empty neighbor pixel group unchanged.

[0051] For example, such as Figure 5As shown, assuming there is a detection image with 4 feature channels and 9 multi-channel pixels, we can sequentially perform four-way value transfer processing on each multi-channel pixel in the detection image to obtain a value transfer feature map corresponding to the detection image. Specifically, taking pixel 5 as an example, the feature channels corresponding to pixel 5 can be divided into upward positive traffic channels, downward positive traffic channels, left positive traffic channels, and right positive traffic channels. Then, since the upward neighbor pixels corresponding to the upward positive traffic channel, the downward neighbor pixels corresponding to the downward positive traffic channel, the left neighbor pixels corresponding to the left positive traffic channel, and the right neighbor pixels corresponding to the right positive traffic channel are all non-empty, we can replace the pixel values ​​in the corresponding upward positive traffic channels with the upward neighbor pixel values, replace the pixel values ​​in the corresponding downward positive traffic channels with the downward neighbor pixel values, replace the pixel values ​​in the corresponding left positive traffic channels with the left neighbor pixel values, and replace the pixel values ​​in the corresponding right positive traffic channels with the right neighbor pixel values.

[0052] The advantage of this setup is that by injecting the values ​​of the neighboring pixels directly adjacent to each multi-channel pixel into itself, it can provide the edge prediction model with a fine-grained and computationally inexpensive local spatial prior, thereby accurately capturing edge details without increasing the burden of additional computing power.

[0053] Optionally, in order to achieve dynamic adaptive fusion of information, a learnable gating vector of the same length as the number of feature channels can be used to weight and fuse the original features of each multi-channel pixel with the features of its neighbors. This allows the edge prediction model to learn the optimal ratio of retaining its own information and incorporating neighborhood information for each feature channel during training.

[0054] Taking an image to be detected with 12 feature channels and a channel segmentation ratio of 3:3:3:3 as an example, the channel segmentation ratio can be adjusted to 5:2:2:3 by using a gating vector, thereby determining that the upward positive traffic channel group, downward positive traffic channel group, left positive traffic channel group, and right positive traffic channel group of the image to be detected include 5, 2, 2, and 3 feature channels, respectively.

[0055] S230. Through the gradient extraction path, perform center difference gradient extraction on the image to be detected to obtain the gradient feature map corresponding to the image to be detected.

[0056] In this step, specifically, such as Figure 4 As shown, a gradient extraction method based on center difference can be used in the gradient extraction path to extract the gradient of the image to be detected by center difference gradient extraction, thereby obtaining the gradient feature map corresponding to the image to be detected.

[0057] For example, in the gradient extraction path, a 3×3 central difference convolution gradient extraction operation can be performed on the image to be detected to obtain the gradient feature map corresponding to the image to be detected.

[0058] The advantage of this setup is that by generating a gradient feature map corresponding to the image to be detected, the local gradient information of the image to be detected can be modeled, thereby enhancing the sensitivity to edges, textures and minor defects.

[0059] S240. Using a context-aware path, extract the context of the image to be detected to obtain a context feature map corresponding to the image to be detected.

[0060] In this step, specifically, such as Figure 4 As shown, in the context-aware path, a context extraction operation based on a large-scale convolutional kernel can be performed on the image to be detected to obtain a context feature map corresponding to the image to be detected. Here, a large-size convolutional kernel can be understood as a convolutional kernel with a side length larger than the size of a regular basic convolutional kernel, such as a 5x5 dilated convolutional kernel.

[0061] With the above settings, more contextual information can be captured to suppress isolated noise points and understand the continuity of the contour.

[0062] S250. The value transfer feature map, gradient feature map and context feature map are fused by feature weighting to obtain the feature enhancement map corresponding to the image to be detected.

[0063] In this step, specifically, such as Figure 4 As shown, the first feature fusion unit in the multidimensional information fusion module can perform feature weighted fusion of the value transfer feature map, gradient feature map and context feature map based on a lightweight, input-driven dynamic gating mechanism to obtain a feature enhancement map corresponding to the image to be detected.

[0064] S260. Based on the feature enhancement map, generate a spatial preprocessing map corresponding to the image to be detected.

[0065] In this step, specifically, the feature enhancement map can be processed through the fully connected layer, nonlinear activation unit, and global information mixer in the multidimensional information fusion module to obtain a spatial preprocessed map corresponding to the image to be detected.

[0066] Optionally, based on the feature enhancement map, a spatial preprocessing map corresponding to the image to be detected is generated, including: processing the feature enhancement map through three fully connected layers respectively to obtain a first fully connected feature map, a second fully connected feature map, and a third fully connected feature map corresponding to the feature enhancement map; performing nonlinear activation on the first fully connected feature map to obtain a linear activation feature map, and performing global information mixing processing on the second and third fully connected feature maps to obtain a global information mixing map; and performing element-wise multiplication on the linear activation feature map and the global information mixing map to obtain the spatial preprocessing map corresponding to the image to be detected.

[0067] Specifically, such as Figure 4 As shown, the feature enhancement map can be processed through a first fully connected layer, a second fully connected layer, and a third fully connected layer to obtain the corresponding first fully connected feature map, second fully connected feature map, and third fully connected feature map. Then, a nonlinear activation unit can be used to nonlinearly activate the first fully connected feature map to obtain a linearly activated feature map. A global information mixer is then used to perform global information mixing processing on the second and third fully connected feature maps to obtain a global information mixture map. Finally, an element-wise multiplication unit can be used to multiply the linearly activated feature map and the global information mixture map element-wise to obtain the spatial preprocessed map corresponding to the image to be detected.

[0068] Furthermore, to illustrate the generation process of the global information hybrid graph in detail, we can use the second fully connected feature map and the third fully connected feature map as examples. Let's take a sequence of pixels as an example. Specifically, the above-mentioned first pixel can be determined using the following specific calculation formula. Output vector of the sequence pixels: .

[0069] in, This indicates the final output number. A vector at each position, The sequence element index represents the position of the currently processed sequence element; in this embodiment, the sequence element specifically refers to a pixel in the image. The position indicates the first 1 pixel, Indicates the first Global interaction features of each pixel; Let be the key vector, given by the first... The input context features of each sequence element are obtained by linear transformation through the second fully connected layer. Let be a value vector, from the first The input context features of each sequence element are obtained by linear transformation through the third fully connected layer. This is the position offset vector; This is the forward molecular state vector; This is the backward molecular state vector; The forward denominator is the state vector; The backward denominator is the state vector; It is a natural constant, approximately 2.718.

[0070] The aforementioned position bias vector is determined during the training process. At the very beginning of training, the value of the position bias vector is initialized, and it is continuously optimized as training progresses until the final value is determined at the end of training. The forward numerator state vector, backward numerator state vector, forward denominator state vector, and backward denominator state vector are continuously updated during model inference. This application obtains these updated vectors according to the following formula: Current state vector = (previous state vector × forget coefficient) + (features of the current pixel).

[0071] Specifically, if the current state vector is the forward numerator state vector of the current pixel, then the corresponding previous state vector is the previous forward numerator state vector of the current pixel; if the current state vector is the backward numerator state vector of the current pixel, then the corresponding previous state vector is the previous backward numerator state vector of the current pixel; if the current state vector is the forward denominator state vector of the current pixel, then the corresponding previous state vector is the previous forward denominator state vector of the current pixel; if the current state vector is the backward denominator state vector of the current pixel, then the corresponding previous state vector is the previous backward denominator state vector of the current pixel; the forgetting coefficient ranges from 0 to 1, and its specific value can be customized according to the actual situation.

[0072] Then, a global information fusion map can be generated based on the output vectors of each sequence of pixels.

[0073] S270. Through the feature compression path, the spatial preprocessed map is refined and reconstructed to obtain the feature compressed map corresponding to the spatial preprocessed map.

[0074] In this step, specifically, to explain the dual-channel entropy change feedforward module in this embodiment in detail, we can first describe the overall structure of the dual-channel entropy change feedforward module: such as Figure 6 As shown, the dual-channel entropy change feedforward module includes a feature compression path, a feature fidelity path, and a second feature fusion unit, and the feature compression path and the feature fidelity path are respectively connected to the second feature fusion unit.

[0075] based on Figure 6The overall structure shown allows for the acquisition of a compressed-size feedforward network with compressed hidden layer dimensions within the feature compression path. This compressed-size feedforward network can be understood as a feedforward network obtained by compressing the width and depth of a standard-size feedforward network; for example, a standard-size feedforward network with its width and depth simultaneously halved can be used as the compressed-size feedforward network. Then, the compressed-size feedforward network can be used to extract and reconstruct features from the spatial preprocessed map in a low-dimensional space, resulting in a compressed feature map corresponding to the spatial preprocessed map.

[0076] The advantage of this setup is that, since the feature compression path compresses both the width and depth of the standard-sized feedforward network, by generating a feature compression map corresponding to the spatial preprocessing map, it is possible to learn a more information-dense and essential core feature representation with almost no impact on image coding efficiency.

[0077] S280. Through the feature fidelity path, a nonlinear transformation is performed on the spatial preprocessed map to obtain the feature fidelity map corresponding to the spatial preprocessed map.

[0078] In this step, specifically, based on Figure 6 The overall structure shown allows for the acquisition of a standard-sized feedforward network within the feature fidelity path. Then, using this standard-sized feedforward network, a nonlinear transformation is performed on the spatial preprocessed map in high-dimensional space to obtain the feature fidelity map corresponding to the spatial preprocessed map.

[0079] By using the above settings, the image edge detection accuracy can be significantly improved while retaining the full expressive power of a standard-sized feedforward network.

[0080] S290. Perform feature fusion on the feature compression map and the feature fidelity map to obtain the encoded feature map corresponding to the spatial preprocessing map.

[0081] In this step, specifically, the feature compression map and the feature fidelity map can be fused by arithmetic averaging to obtain the encoded feature map corresponding to the spatial preprocessing map.

[0082] The advantage of this setup is that by introducing a random path selection mechanism during the training phase, the network is forced to learn complementary and robust feature representations, and "zero-cost" model integration is achieved during the inference phase, which can significantly improve the model's generalization ability and detection accuracy in complex industrial environments.

[0083] S2100: The decoder upsamples the encoded feature map to obtain the edge prediction image corresponding to the image to be detected.

[0084] The technical solution of this embodiment, by constructing a multi-dimensional information fusion module capable of parallel processing and adaptively fusing multi-dimensional information such as neighboring pixel values, gradients, and context, fundamentally enhances the edge prediction model's ability to finely understand the local structure of images, thereby accurately capturing weak or blurred edge details in images. Secondly, by introducing a random path selection mechanism during the training phase, forcing the network to learn complementary and robust feature representations, and achieving "zero-cost" model integration during the inference phase, the edge prediction model's resistance to common industrial interferences such as illumination changes, electromagnetic noise, and image blurring can be significantly improved, ensuring the stability and reliability of edge detection results. Finally, by improving the existing VRWKV model to obtain an edge prediction model, and using this model for image edge prediction, the efficiency bottleneck of the existing ViT model when processing high-resolution images can be fundamentally solved, perfectly meeting the stringent requirements of the nuclear power industry for high-reliability and high-precision online detection.

[0085] Example 3 Figure 7 This is a schematic diagram of a multidimensional collaborative adaptive image edge detection device according to Embodiment 3 of the present invention. This embodiment is applicable to the case of edge detection of images. The multidimensional collaborative adaptive image edge detection device can be implemented in hardware and / or software and can be configured in electronic devices such as computers.

[0086] like Figure 7 As shown, the multidimensional collaborative adaptive image edge detection device disclosed in this embodiment includes: an information acquisition module 71, a spatial preprocessing module 72, an image encoding module 73, and an image decoding module 74.

[0087] The information acquisition module 71 is used to acquire the image to be detected and the pre-built edge prediction model. The edge prediction model includes an encoder and a decoder. The encoder includes a multi-dimensional information fusion module and a dual-channel entropy change feedforward module.

[0088] The spatial preprocessing module 72 is used to determine the value transfer feature map, gradient feature map and context feature map corresponding to the image to be detected in parallel through the multi-dimensional information fusion module, and generate a spatial preprocessing map corresponding to the image to be detected based on the value transfer feature map, gradient feature map and context feature map.

[0089] The image encoding module 73 is used to determine the feature compression map and feature fidelity map corresponding to the spatial preprocessing map in parallel through the dual-channel entropy change feedforward module, and to perform feature fusion on the feature compression map and feature fidelity map to obtain the encoded feature map corresponding to the spatial preprocessing map.

[0090] The image decoding module 74 is used to upsample the encoded feature map through the decoder to obtain the edge prediction image corresponding to the image to be detected.

[0091] The technical solution in this embodiment, through the cooperation of the information acquisition module 71, the spatial preprocessing module 72, the image encoding module 73, and the image decoding module 74, solves the problem that it is difficult to simultaneously take into account the three performance indicators of efficiency, accuracy, and robustness when using the ViT model or VRWKV model for image edge detection, and significantly improves the efficiency, accuracy, and robustness of image edge detection.

[0092] Optionally, when the multidimensional information fusion module includes a value transfer path, a gradient extraction path, and a context-aware path, the spatial preprocessing module 72 includes: a value transfer feature map generation unit, a gradient feature map generation unit, a context feature map generation unit, a feature map fusion unit, and a spatial preprocessing map generation unit.

[0093] The value transfer feature map generation unit is used to obtain each multi-channel pixel in the image to be detected and the corresponding neighboring pixel group in each multi-channel pixel through the value transfer path, and transfer the neighboring pixel value in each neighboring pixel group to the corresponding multi-channel pixel to obtain the value transfer feature map corresponding to the image to be detected.

[0094] The gradient feature map generation unit is used to extract the central difference gradient of the image to be detected through the gradient extraction path, so as to obtain the gradient feature map corresponding to the image to be detected.

[0095] The context feature map generation unit is used to extract the context of the image to be detected through a context-aware path, and obtain the context feature map corresponding to the image to be detected.

[0096] The feature map fusion unit is used to perform feature weighted fusion of the value transfer feature map, gradient feature map and context feature map to obtain the feature enhancement map corresponding to the image to be detected.

[0097] The spatial preprocessing map generation unit is used to generate a spatial preprocessing map corresponding to the image to be detected based on the feature enhancement map.

[0098] Optionally, the value transfer feature map generation unit is specifically used for: obtaining each multi-channel pixel in the image to be detected through the value transfer path, and determining the upward positive traffic channel group, downward positive traffic channel group, left positive traffic channel group, and right positive traffic channel group corresponding to each multi-channel pixel based on the number of feature channels corresponding to the image to be detected; determining the upward neighbor pixel group, downward neighbor pixel group, left neighbor pixel group, and right neighbor pixel group corresponding to each multi-channel pixel based on the upward positive traffic channel group, downward positive traffic channel group, left positive traffic channel group, and right positive traffic channel group corresponding to each multi-channel pixel; replacing the pixel values ​​in the upward positive traffic channel group, downward positive traffic channel group, left positive traffic channel group, and right positive traffic channel group corresponding to each multi-channel pixel with the upward neighbor pixel group, downward neighbor pixel group, left neighbor pixel group, and right neighbor pixel group corresponding to each multi-channel pixel to obtain the value transfer feature map corresponding to the image to be detected.

[0099] Optionally, the spatial preprocessing map generation unit is specifically used to: process the feature enhancement map through three fully connected layers respectively to obtain a first fully connected feature map, a second fully connected feature map, and a third fully connected feature map corresponding to the feature enhancement map; perform nonlinear activation on the first fully connected feature map to obtain a linear activation feature map, and perform global information mixing processing on the second and third fully connected feature maps to obtain a global information mixing map; and perform element-wise multiplication on the linear activation feature map and the global information mixing map to obtain a spatial preprocessing map corresponding to the image to be detected.

[0100] Optionally, when the dual-channel entropy-change feedforward module includes a feature compression path and a feature fidelity path, the image encoding module 73 is specifically used to: extract and reconstruct features from the spatial preprocessed image through the feature compression path to obtain a feature compression image corresponding to the spatial preprocessed image; and perform a nonlinear transformation on the spatial preprocessed image through the feature fidelity path to obtain a feature fidelity image corresponding to the spatial preprocessed image.

[0101] Optionally, the device further includes a model training module, which is used to: acquire training images and an edge prediction model to be trained; wherein the edge prediction model to be trained includes an encoder to be trained and a decoder to be trained, and the encoder to be trained includes a multi-dimensional information fusion module to be trained and a dual-channel entropy-change feedforward module to be trained; generate a spatial preprocessing map corresponding to the training image through the multi-dimensional information fusion module to be trained; perform feature compression or feature fidelity operation on the spatial preprocessing map through the dual-channel entropy-change feedforward module to be trained, and determine the feature encoding representation map corresponding to the training image based on the feature compression or feature fidelity result; perform upsampling processing on the feature encoding representation map through the decoder to be trained to obtain an edge prediction image corresponding to the training image; determine whether the edge detection accuracy of the edge prediction model to be trained meets the preset accuracy requirements based on each training image and the edge prediction image corresponding to each training image; if not, continue to train the edge prediction model to be trained until a trained edge prediction model is obtained.

[0102] The multidimensional collaborative adaptive image edge detection device provided in this embodiment of the invention can execute the multidimensional collaborative adaptive image edge detection method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Content not described in detail in this embodiment can be referred to the description in any method embodiment of this application.

[0103] Example 4 Figure 8 A schematic diagram of the structure of an electronic device 10 that can be used to implement embodiments of the present invention is shown. For example... Figure 8 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory 12 or a random access memory 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the read-only memory 12 or loaded from storage unit 18 into the random access memory 13. The random access memory 13 may also store various programs and data required for the operation of the electronic device 10. The processor 11, read-only memory 12, and random access memory 13 are interconnected via a bus 14. An input / output interface 15 is also connected to the bus 14.

[0104] Multiple components in electronic device 10 are connected to input / output interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of monitors, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0105] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as multidimensional collaborative adaptive image edge detection methods.

[0106] In some embodiments, the multidimensional collaborative adaptive image edge detection method can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via read-only memory 12 and / or communication unit 19. When the computer program is loaded into random access memory 13 and executed by processor 11, one or more steps of the multidimensional collaborative adaptive image edge detection method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the multidimensional collaborative adaptive image edge detection method by any other suitable means (e.g., by means of firmware).

[0107] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0108] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0109] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0110] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0111] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0112] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0113] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0114] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A multidimensional collaborative adaptive image edge detection method, characterized in that, The multidimensional collaborative adaptive image edge detection method includes: The image to be detected and a pre-constructed edge prediction model are acquired; wherein the edge prediction model includes an encoder and a decoder, and the encoder includes a multi-dimensional information fusion module and a dual-channel entropy change feedforward module. The multidimensional information fusion module determines the value transfer feature map, gradient feature map, and context feature map corresponding to the image to be detected in parallel, and generates a spatial preprocessing map corresponding to the image to be detected based on the value transfer feature map, gradient feature map, and context feature map; the multidimensional information fusion module includes a value transfer path; The dual-channel entropy-change feedforward module determines the feature compression map and feature fidelity map corresponding to the spatial preprocessing map in parallel, and performs feature fusion on the feature compression map and feature fidelity map to obtain the encoded feature map corresponding to the spatial preprocessing map. The decoder upsamples the encoded feature map to obtain an edge prediction image corresponding to the image to be detected. The determination of the value transfer feature map corresponding to the image to be detected through the multi-dimensional information fusion module includes: The multi-channel pixels in the image to be detected are obtained through the value transmission path, and based on the number of feature channels corresponding to the image to be detected, the upward positive traffic channel group, downward positive traffic channel group, left positive traffic channel group and right positive traffic channel group corresponding to each multi-channel pixel are determined respectively. Based on the upward positive traffic channel group, downward positive traffic channel group, left positive traffic channel group and right positive traffic channel group corresponding to each of the multi-channel pixels, the upward neighbor pixel group, downward neighbor pixel group, left neighbor pixel group and right neighbor pixel group corresponding to each of the multi-channel pixels are determined respectively. The pixel values ​​in the upward, downward, left, and right neighboring pixel groups corresponding to each of the multi-channel pixels are replaced by the pixel values ​​in the upward, downward, left, and right forward traffic channels corresponding to each of the multi-channel pixels, to obtain the value transfer feature map corresponding to the image to be detected.

2. The multidimensional collaborative adaptive image edge detection method according to claim 1, characterized in that, The multidimensional information fusion module further includes a gradient extraction path and a context-aware path. Through this module, the gradient feature map and context feature map corresponding to the image to be detected are determined in parallel, including: The center difference gradient is extracted from the image to be detected through the gradient extraction path to obtain a gradient feature map corresponding to the image to be detected. The context is extracted from the image to be detected through the context-aware path to obtain a context feature map corresponding to the image to be detected.

3. The multidimensional collaborative adaptive image edge detection method according to claim 1, characterized in that, Based on the value transfer feature map, gradient feature map, and context feature map, a spatial preprocessing map corresponding to the image to be detected is generated, including: The value transfer feature map, gradient feature map, and context feature map are fused by feature weighting to obtain a feature enhancement map corresponding to the image to be detected. Based on the feature enhancement map, a spatial preprocessing map corresponding to the image to be detected is generated.

4. The multidimensional collaborative adaptive image edge detection method according to claim 3, characterized in that, Based on the feature enhancement map, a spatial preprocessing map corresponding to the image to be detected is generated, including: The feature enhancement map is processed through three fully connected layers to obtain a first fully connected feature map, a second fully connected feature map, and a third fully connected feature map corresponding to the feature enhancement map. The first fully connected feature map is nonlinearly activated to obtain a nonlinearly activated feature map, and the second fully connected feature map and the third fully connected feature map are subjected to global information mixing processing to obtain a global information mixing map. The nonlinear activation feature map and the global information mixture map are multiplied element-wise to obtain a spatial preprocessing map corresponding to the image to be detected.

5. The multidimensional collaborative adaptive image edge detection method according to claim 1, characterized in that, The dual-channel entropy-change feedforward module includes a feature compression path and a feature fidelity path. Through this module, the feature compression map and feature fidelity map corresponding to the spatial preprocessed map are determined in parallel, including: The spatial preprocessed map is processed and reconstructed using the feature compression path to obtain a feature compressed map corresponding to the spatial preprocessed map. The spatial preprocessed map is nonlinearly transformed using the feature fidelity path to obtain a feature fidelity map corresponding to the spatial preprocessed map.

6. The multidimensional collaborative adaptive image edge detection method according to claim 5, characterized in that, Before obtaining the pre-built edge prediction model, the following is also included: Acquire training images and an edge prediction model to be trained; wherein the edge prediction model to be trained includes an encoder to be trained and a decoder to be trained, and the encoder to be trained includes a multi-dimensional information fusion module to be trained and a dual-channel entropy change feedforward module to be trained. The multidimensional information fusion module to be trained generates a spatial preprocessing map corresponding to the training image. The dual-channel entropy-change feedforward module to be trained performs feature compression or feature fidelity operation on the spatial preprocessing image, and determines the feature encoding representation map corresponding to the training image based on the feature compression or feature fidelity result. The feature encoding representation map is upsampled using the decoder to be trained to obtain an edge prediction image corresponding to the training image. Based on each of the training images and the edge prediction images corresponding to each of the training images, determine whether the edge detection accuracy of the edge prediction model to be trained meets the preset accuracy requirements; If not, continue training the edge prediction model to be trained until a fully trained edge prediction model is obtained.

7. A multidimensional collaborative adaptive image edge detection device, characterized in that, The multidimensional collaborative adaptive image edge detection device includes: An information acquisition module is used to acquire the image to be detected and a pre-built edge prediction model; wherein, the edge prediction model includes an encoder and a decoder, and the encoder includes a multi-dimensional information fusion module and a dual-channel entropy change feedforward module; The spatial preprocessing module is used to determine, in parallel, the value transfer feature map, gradient feature map, and context feature map corresponding to the image to be detected through the multi-dimensional information fusion module, and to generate a spatial preprocessing map corresponding to the image to be detected based on the value transfer feature map, gradient feature map, and context feature map; the multi-dimensional information fusion module includes a value transfer path; The image encoding module is used to determine, in parallel, the feature compression map and the feature fidelity map corresponding to the spatial preprocessing map through the dual-channel entropy change feedforward module, and to perform feature fusion on the feature compression map and the feature fidelity map to obtain the encoded feature map corresponding to the spatial preprocessing map. The image decoding module is used to upsample the encoded feature map through the decoder to obtain an edge prediction image corresponding to the image to be detected; The spatial preprocessing module includes a value-transfer feature map generation unit, which is specifically used for: The multi-channel pixels in the image to be detected are obtained through the value transmission path, and based on the number of feature channels corresponding to the image to be detected, the upward positive traffic channel group, downward positive traffic channel group, left positive traffic channel group and right positive traffic channel group corresponding to each multi-channel pixel are determined respectively. Based on the upward positive traffic channel group, downward positive traffic channel group, left positive traffic channel group and right positive traffic channel group corresponding to each of the multi-channel pixels, the upward neighbor pixel group, downward neighbor pixel group, left neighbor pixel group and right neighbor pixel group corresponding to each of the multi-channel pixels are determined respectively. The pixel values ​​in the upward, downward, left, and right neighboring pixel groups corresponding to each of the multi-channel pixels are replaced by the pixel values ​​in the upward, downward, left, and right forward traffic channels corresponding to each of the multi-channel pixels, to obtain the value transfer feature map corresponding to the image to be detected.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the multidimensional collaborative adaptive image edge detection method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the multidimensional collaborative adaptive image edge detection method according to any one of claims 1-6.

Citation Information

Patent Citations

  • Image edge detection method based on deep supervision mask self-encoding network

    CN118334065A

  • Image adaptive feature enhancement and layered attention method for forest fire detection

    CN121789089A