Image panorama segmentation method and device based on feature enhancement, equipment and storage medium
By employing a feature-enhanced image panorama segmentation method, which utilizes multi-level feature extraction and redistribution, combined with attention mechanisms and convolutional kernel generation, the problem of insufficient accuracy in existing image panorama segmentation technologies is solved, achieving higher scene understanding and target recognition accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- MACAO POLYTECHNIC INST
- Filing Date
- 2024-03-06
- Publication Date
- 2026-05-12
AI Technical Summary
Existing technologies lack effective attention mechanisms and coordinate convolution in panoramic image segmentation, resulting in decreased performance when processing small objects or scenes with rich details.
A feature-enhanced image panorama segmentation method is adopted, including preprocessing, multi-level feature extraction, feature enhancement, target feature redistribution, and panorama segmentation. It utilizes the strong generalization feature generation module, stage feature redistribution module, and ensemble encoder module in the image processing network, and improves the segmentation accuracy through feature fusion, convolutional kernel generation, and attention mechanisms.
It improves the understanding of activities and events in image scenes, increases the accuracy of target object detection and recognition, and enhances the precision of panoramic segmentation.
Smart Images
Figure CN117975222B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer vision recognition technology, and in particular to a method, apparatus, device and storage medium for panoramic image segmentation based on feature enhancement. Background Technology
[0002] With the continuous advancement of computer vision research and deep learning methods, deep learning-based techniques such as image classification, semantic segmentation, and instance segmentation have made significant progress. Semantic segmentation assigns a semantic category label to each pixel in an image, but it cannot distinguish between different object instances of the same semantic category. Instance segmentation performs pixel-level segmentation of object instances in an image, but it does not involve various uncountable objects without definite shapes. Panoptic segmentation is a unification of semantic segmentation and instance segmentation tasks, and it is crucial for tasks that rely on visual perception of image scenes, such as autonomous driving and intelligent robotics.
[0003] However, existing technologies may not effectively apply attention mechanisms and coordinate convolution, resulting in shortcomings in redistributing large amounts of channel information and providing accurate spatial information, leading to performance degradation when processing small objects or scenes with rich details.
[0004] In view of this, there is an urgent need for an RGB image panoramic segmentation prediction optimization method that employs multiple feature enhancement techniques for use in autonomous driving and video surveillance, thereby improving the understanding of activities and events in the scene and better realizing the detection and recognition of target objects.
[0005] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention
[0006] The main objective of this invention is to provide a feature-enhanced image panoramic segmentation method, apparatus, device, and storage medium, aiming to solve the technical problem of low accuracy in existing image panoramic segmentation techniques.
[0007] To achieve the above objectives, the present invention provides a feature-enhanced image panoramic segmentation method, the method comprising the following steps:
[0008] The original image is preprocessed to obtain the processed original image;
[0009] The processed original image is input into an image processing network for image feature extraction to obtain a multi-level feature image set.
[0010] Image feature enhancement is performed based on the multi-level feature image set to obtain an enhanced feature image set.
[0011] The enhanced feature image set is then redistributed to obtain the target feature image set.
[0012] The target feature image set is subjected to panoramic segmentation processing to obtain a panoramic segmented image.
[0013] Optionally, image feature enhancement is performed based on the multi-level feature image set to obtain an enhanced feature image set, specifically including:
[0014] The feature images at different levels in the multi-level feature image set are fused according to their scale to obtain a fused feature image set.
[0015] Image feature enhancement is performed on the fused feature image set to obtain the enhanced feature image set.
[0016] Optionally, after performing image feature enhancement based on the multi-level feature image set to obtain the enhanced feature image set, the method further includes:
[0017] Based on the enhanced feature image set, a convolution kernel is generated to obtain the panoramic image segmentation convolution kernel.
[0018] Optionally, the enhanced feature image set is subjected to target feature reallocation to obtain a target feature image set, specifically including:
[0019] The enhanced feature image set is then merged to obtain the overall stage feature image;
[0020] Target feature weights are calculated based on the enhanced feature image set to obtain stage attention weight data.
[0021] The target feature image set is obtained by integrating the overall stage feature image with the stage attention weight data.
[0022] Optionally, the target feature image set is subjected to panoramic segmentation processing to obtain a panoramic segmentation image, specifically including:
[0023] The target feature image set is integrated and encoded to obtain the encoded target feature image;
[0024] The encoded target feature image is decoded to obtain the panoramic segmentation result;
[0025] The panoramic segmentation result is used to perform image post-processing to obtain the panoramic segmentation image.
[0026] Optionally, the target feature image set is integrated and encoded to obtain the encoded target feature image, specifically including:
[0027] Encoded feature extraction is performed on the target feature image set to obtain spatial coded feature images;
[0028] Attention weights are redistributed based on the spatial coding feature image to obtain a unified coding feature image;
[0029] Based on the panoramic segmentation convolutional kernel of the image, the unified coded feature image is segmented and predicted to obtain the coded target feature image.
[0030] Optionally, the encoded target feature image is decoded to obtain a panoramic segmentation result, specifically including:
[0031] The encoded target feature image is upsampled to obtain a sampled target feature image;
[0032] The sampled target feature image is subjected to feature fusion processing to obtain the panoramic segmentation result.
[0033] Furthermore, to achieve the above objectives, the present invention also proposes a feature-enhanced image panorama segmentation device, wherein the feature-enhanced image panorama segmentation device comprises:
[0034] Preprocessing module: preprocesses the original image to obtain the processed original image;
[0035] Feature extraction module: Inputs the processed original image into an image processing network to extract image features and obtain a multi-level feature image set;
[0036] Feature enhancement module: Performs image feature enhancement based on the multi-level feature image set to obtain an enhanced feature image set;
[0037] Feature reassignment module: performs target feature reassignment on the enhanced feature image set to obtain a target feature image set;
[0038] Panoramic segmentation module: Performs panoramic segmentation processing on the target feature image set to obtain a panoramic segmented image.
[0039] Furthermore, to achieve the above objectives, the present invention also proposes a feature-enhanced image panorama segmentation device, which includes: a memory, a processor, and a feature-enhanced image panorama segmentation program stored in the memory and executable on the processor. The feature-enhanced image panorama segmentation program is configured to implement the steps of the feature-enhanced image panorama segmentation method described above.
[0040] Furthermore, to achieve the above objectives, the present invention also proposes a computer-readable storage medium storing a computer program, wherein the storage medium stores a feature-enhanced image panorama segmentation program, and when the feature-enhanced image panorama segmentation program is executed by a processor, it implements the steps of the feature-enhanced image panorama segmentation method described above.
[0041] This invention preprocesses the original image to obtain a processed original image; it then inputs the processed original image into an image processing network for image feature extraction, resulting in a multi-level feature image set; based on the multi-level feature image set, it performs image feature enhancement to obtain an enhanced feature image set; it then performs target feature redistribution on the enhanced feature image set to obtain a target feature image set; finally, it performs panoramic segmentation processing on the target feature image set to obtain a panoramic segmented image. This invention inputs the processed original image into an image processing network for sequential feature extraction, feature enhancement, and target feature redistribution, ultimately generating a panoramic segmented image corresponding to the original image. This improves the understanding of activities and events in the image scene and increases the accuracy of target object detection and recognition. Attached Figure Description
[0042] Figure 1 This is a schematic diagram of the structure of a feature-enhanced image panoramic segmentation device in the hardware operating environment involved in the embodiments of the present invention;
[0043] Figure 2 This is a flowchart illustrating the first embodiment of the feature-enhanced image panoramic segmentation method of the present invention;
[0044] Figure 3 This is a flowchart illustrating the second embodiment of the feature-enhanced image panoramic segmentation method of the present invention;
[0045] Figure 4 This is a flowchart illustrating the third embodiment of the feature-enhanced image panoramic segmentation method of the present invention;
[0046] Figure 5 This is a flowchart illustrating the fourth embodiment of the feature-enhanced image panoramic segmentation method of the present invention;
[0047] Figure 6 This is a structural block diagram of the first embodiment of the image panoramic segmentation device based on feature enhancement of the present invention.
[0048] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0049] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.
[0050] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of a feature-enhanced image panoramic segmentation device in the hardware operating environment of the embodiment of the present invention.
[0051] like Figure 1 As shown, the feature-enhanced image panoramic segmentation device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.
[0052] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on feature-enhanced image panoramic segmentation devices and may include more or fewer components than shown, or combine certain components, or have different component arrangements.
[0053] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a feature-enhanced image panoramic segmentation program.
[0054] exist Figure 1 In the feature-enhanced image panoramic segmentation device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and memory 1005 in the feature-enhanced image panoramic segmentation device of the present invention can be set in the feature-enhanced image panoramic segmentation device, and the feature-enhanced image panoramic segmentation device calls the feature-enhanced image panoramic segmentation program stored in the memory 1005 through the processor 1001 and executes the feature-enhanced image panoramic segmentation method provided in the embodiment of the present invention.
[0055] This invention provides a feature-enhanced image panoramic segmentation method, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of a feature-enhanced image panoramic segmentation method according to the present invention.
[0056] In this embodiment, the feature-enhanced image panoramic segmentation method includes the following steps:
[0057] Step S10: Preprocess the original image to obtain the processed original image;
[0058] It should be noted that, in the specific implementation, the preprocessing of the original image includes normalization and image size adjustment, the purpose of which is to make the processed original image adaptable to the input requirements of the image processing network.
[0059] Step S20: Input the processed original image into the image processing network for image feature extraction to obtain a multi-level feature image set;
[0060] It is understandable that, in the specific implementation, the image feature extraction of the processed original image is achieved through the basic image pyramid fully convolutional network in the image processing network; and the various feature images in the resulting multi-level feature image set include multi-scale spatial information from shallow to deep layers.
[0061] Step S30: Perform image feature enhancement based on the multi-level feature image set to obtain the enhanced feature image set;
[0062] It should be noted that the specific process of image feature enhancement is as follows: first, feature images of different scales in a multi-level feature image set are fused separately using a multi-scale feature generation method; then, a deformable convolutional network is used to enhance the connections between the fused feature images, ultimately generating multiple image features at different resolutions. Its purpose is to improve the image processing network's ability to recognize objects of different sizes in the image.
[0063] It should also be noted that, in the specific implementation, image feature enhancement is based on the strong generalization feature generation module in the image processing network.
[0064] Step S40: Redistribute target features on the enhanced feature image set to obtain the target feature image set;
[0065] It should be noted that, in the specific implementation, the target feature reassignment of the enhanced feature image set is based on the stage feature reassignment module in the image processing network. When the attention module in the stage feature reassignment module performs stage reassignment of feature images, it enables the image processing network to learn the importance of features at different resolutions and emphasize the expression of features at important resolutions, while suppressing unimportant features, thereby improving the representation efficiency of feature images.
[0066] Step S50: Perform panoramic segmentation processing on the target feature image set to obtain a panoramic segmentation image.
[0067] It should be noted that, in the specific implementation, the panoramic segmentation process includes integrated encoding, decoding output, and post-processing. The specific process is as follows: First, the target feature image set is input into the integrated encoder module of the image processing network, which uses coordinate convolution to inject spatial information into the feature image, thereby improving the accuracy of boundary localization; then, the encoded feature image is upsampled and further fused by the decoder to output a refined panoramic segmentation result; finally, the panoramic segmentation result is post-processed to generate the final panoramic segmentation image.
[0068] It should also be noted that during the decoding output processing, the channel attention mechanism is used to further compress and redistribute channel information, reduce feature dimensions, and improve coding efficiency. The post-processing process specifically includes thresholding, connected component analysis, etc., which are used to adjust and optimize the output panoramic segmentation results to obtain the final panoramic segmentation image.
[0069] This embodiment preprocesses the original image to obtain a processed original image; the processed original image is then input into an image processing network for image feature extraction, resulting in a multi-level feature image set; image feature enhancement is performed based on the multi-level feature image set, resulting in an enhanced feature image set; target feature redistribution is performed on the enhanced feature image set, resulting in a target feature image set; and panoramic segmentation is performed on the target feature image set to obtain a panoramic segmented image. This embodiment inputs the processed original image into an image processing network for sequential feature extraction, feature enhancement, and target feature redistribution, ultimately generating a panoramic segmented image corresponding to the original image. This improves the understanding of activities and events in the image scene and increases the accuracy of target object detection and recognition.
[0070] refer to Figure 3 , Figure 3 This is a flowchart illustrating a second embodiment of a feature-enhanced image panoramic segmentation method according to the present invention.
[0071] Based on the first embodiment described above, in this embodiment, step S30 specifically includes the following steps:
[0072] Step S31: Perform feature fusion on the feature images of different levels in the multi-level feature image set according to their scale to obtain the fused feature image set;
[0073] It should be noted that in the specific implementation, if given an RGB three-channel image that needs to be segmented into a panorama... After image feature extraction, it can generate four backbone features at different resolutions. i∈(2,3,4,5), where C c The number of channels representing features, and these image features at different resolutions constitute a multi-level feature image set.
[0074] It should also be noted that, in the specific implementation, the process of fusing features of different levels in the multi-level feature image set according to their scale is as follows: if the feature images of different levels are named Ci, then each feature image of different levels can be distinguished and named, such as C1, and then multiple features of C1 with different scales can be generated. Then, feature images of different levels but with the same size are fused to obtain a fused feature image set. For example, C1, C2 and C3 with the same size are fused to obtain a fused feature image set.
[0075] Step S32: Perform image feature enhancement on the fused feature image set to obtain the enhanced feature image set.
[0076] Understandably, in the specific implementation, the enhanced feature image set is generated by the strong generalization feature generation module in the image processing network using a multi-scale feature generation method. The specific process can be as follows: first, the convolutional feature (Conv Feature) M4 generated by the strong generalization feature generation module is multisampled, including 2x downsampling and 4x upsampling to the required resolution. Then, it is merged with other features at the same resolution generated by the SCFG (Sampled Concated Feature Generation multi-scale feature generation module) to generate four sets of feature generalization enhancement combinations L2-L5, which can be expressed by the formula:
[0077]
[0078] Here, Concat(·) represents the merging operation, DW(·) represents the downsampling operation, and UP(·) represents the upsampling operation. Specifically, each convolutional feature M iUpsampling will be performed using bilinear interpolation, followed by downsampling using max pooling to generate two additional features. and After that, M i , and The features are then arranged and combined with the other two convolutional features in an L2-L5 manner. Finally, GSFGM generates multi-scale stage features at six different resolutions (P2-P7) through multiple 3×3 convolutional networks, a 3×3 deformable convolutional network (DCN), and a 2x upsampling operation. i∈(2,3,4,5,6,7), the multi-scale stage features obtained at different resolutions together constitute the enhanced feature image set.
[0079] In this embodiment, feature fusion is first performed on a multi-level feature image set, and then image feature enhancement is performed on it to obtain an enhanced feature image set, which improves the accuracy of subsequent panoramic image segmentation.
[0080] Furthermore, after performing image feature enhancement based on the multi-level feature image set to obtain the enhanced feature image set, the method further includes: generating a convolution kernel based on the enhanced feature image set to obtain an image panoramic segmentation convolution kernel.
[0081] It should be noted that, in the specific implementation, it can be based on 5 features from the multi-level feature image set. For i∈(3,4,5,6,7), a unified panoramic segmentation kernel generation method is used to generate panoramic segmentation convolution kernels.
[0082] It should also be noted that the panoptic segmentation convolution kernel generation process includes generating kernels for each P. i The system generates object centers, object regions, and weighted convolutional kernels. These kernels are then fused based on cosine similarity using a clustering method to generate the final panoptic segmentation convolutional kernel. Where M and N represent the number of background and item categories that need to be segmented, respectively.
[0083] refer to Figure 4 , Figure 4 This is a flowchart illustrating a third embodiment of a feature-enhanced image panoramic segmentation method according to the present invention.
[0084] Based on the first embodiment described above, in this embodiment, step S40 specifically includes the following steps:
[0085] Step S41: Merge the enhanced feature image set to obtain the overall stage feature image;
[0086] It should be noted that in the actual implementation, since feature images of different scales have different receptive fields, target feature enhancement is required to preserve more details of objects at different scales.
[0087] It should also be noted that, in the specific implementation, feature image merging is performed on the enhanced feature image set. Specifically, this means merging the four stage features P2-P5 to generate the overall stage feature (CSF), which is the overall stage feature image.
[0088] Step S42: Calculate the target feature weights based on the enhanced feature image set to obtain the stage attention weight data;
[0089] Understandably, in the specific implementation, in order to maintain resolution and retain more feature details, the image processing network first performs upsampling on the feature images P3-P5 using bilinear interpolation and convolution operations, and then applies the stage attention mechanism to reweight the contribution of each stage feature, thereby obtaining the stage attention weight data.
[0090] It should also be noted that the specific process for generating stage attention weight data is as follows: first, a global average pooling operation is applied to transfer the overall stage feature image from C in the stage feature redistribution module. s ×W / 4×H / 4 maps to N s A 1×1 vector is generated, and then two fully connected layers are used to generate the stage attention weights Z = z. s ,s∈(2,3,4,5).
[0091] Step S43: Integrate the overall stage feature image and stage attention weight data to obtain a set of target feature images.
[0092] Understandably, the integration and processing of the overall stage feature image and the stage attention weight data involves multiplying the attention weight Z with the overall stage feature image to obtain the re-weighted stage feature (RSF), which is the set of target feature images.
[0093] In this embodiment, feature images are first merged based on the enhanced feature image set to obtain the overall stage feature image. Then, the obtained stage attention weight data is integrated with the overall stage feature image to finally obtain the target feature image set. This suppresses unimportant features and improves the representation efficiency of the feature images.
[0094] refer to Figure 5 , Figure 5This is a flowchart illustrating the fourth embodiment of a feature-enhanced image panoramic segmentation method according to the present invention.
[0095] Based on the first embodiment described above, in this embodiment, step S50 specifically includes the following steps:
[0096] Step S51: Perform integrated encoding processing on the target feature image set to obtain the encoded target feature image;
[0097] It should be noted that, in the specific implementation, the purpose of integrating and encoding the target feature image set is to further extract and compress the data of the target feature image set, thereby improving the generalization ability of the target feature image set data.
[0098] It should also be noted that, in the specific implementation, the enumeration encoding process involves first adding two additional channels containing x and y coordinate information to the stage features after weight redistribution, and then applying coordinate convolution to obtain the Spatial Encoded Feature (SEF). SEF Finally, the encoded feature S e Perform a convolution operation with the panoptic segmentation kernel K to obtain all classification prediction instances P, i.e., Pi i =Conv(K j ,S e ), where P i Let K represent a prediction instance. j S represents the corresponding panoptic segmentation convolution kernel. e This represents the unified encoding feature, and Conv(·,·) represents the convolution operation.
[0099] Understandably, in the specific implementation, the panoramic segmentation process performed by the integrated encoder module applies a channel attention mechanism, which adaptively reallocates the weights of feature channels based on the correlation between different channels. Specifically, it first uses global average pooling to compress the SEF into C... e A ×1×1 vector; then C is generated through two fully connected layers. e The SEF channel attention weight Z = z ×1×1 c ,c∈(1,C e ), where z c Representing SEF Attention weights for specific channels.
[0100] Step S52: Decode the encoded target feature image to obtain the panoramic segmentation result;
[0101] It should be noted that in the specific implementation, the encoded target feature image can be upsampled and further fused by the decoder to finally output the corresponding panoramic segmentation result. The decoder may include transposed convolutional layers or upsampling layers, as well as other possible feature fusion techniques, such as skip connections, to preserve high-resolution detail information.
[0102] Step S53: Perform image post-processing based on the panoramic segmentation results to obtain the panoramic segmentation image.
[0103] It is understandable that the image post-processing process includes thresholding, connected component analysis, etc., which are used to adjust and optimize the output panoramic segmentation results to obtain the final panoramic segmentation image.
[0104] In this embodiment, the target feature image set is first integrated and encoded, and then the encoded target feature image is decoded to obtain the corresponding panoramic segmentation result. Finally, the panoramic segmentation image is obtained through post-processing, thus realizing the detection and recognition of target objects in the original image.
[0105] Further, the target feature image set is integrated and encoded to obtain the encoded target feature image, specifically including: extracting encoded features from the target feature image set to obtain a spatial encoded feature image; redistributing attention weights based on the spatial encoded feature image to obtain a unified encoded feature image; and performing segmentation prediction on the unified encoded feature image based on the image panoramic segmentation convolution kernel to obtain the encoded target feature image.
[0106] Further, the encoded target feature image is decoded to obtain a panoramic segmentation result, specifically including: upsampling the encoded target feature image to obtain a sampled target feature image; and performing feature fusion processing on the sampled target feature image to obtain a panoramic segmentation result.
[0107] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, wherein the storage medium stores a feature-enhanced image panorama segmentation program, and when the feature-enhanced image panorama segmentation program is executed by a processor, it implements the steps of the feature-enhanced image panorama segmentation method described above.
[0108] Since this storage medium adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought about by the technical solutions of the above embodiments, which will not be repeated here.
[0109] Reference Figure 6 , Figure 6 This is a structural block diagram of the first embodiment of the image panoramic segmentation device based on feature enhancement of the present invention.
[0110] like Figure 6 As shown, the image panoramic segmentation device based on feature enhancement proposed in this embodiment of the invention includes:
[0111] Preprocessing module 10: preprocesses the original image to obtain the processed original image;
[0112] Feature extraction module 20: Inputs the processed original image into the image processing network to extract image features and obtain a multi-level feature image set;
[0113] Feature enhancement module 30: Performs image feature enhancement based on a multi-level feature image set to obtain an enhanced feature image set;
[0114] Feature redistribution module 40: redistributes target features to the enhanced feature image set to obtain the target feature image set;
[0115] Panoramic segmentation module 50: Performs panoramic segmentation processing on the target feature image set to obtain a panoramic segmented image.
[0116] This embodiment preprocesses the original image to obtain a processed original image; the processed original image is then input into an image processing network for image feature extraction, resulting in a multi-level feature image set; image feature enhancement is performed based on the multi-level feature image set, resulting in an enhanced feature image set; target feature redistribution is performed on the enhanced feature image set, resulting in a target feature image set; and panoramic segmentation is performed on the target feature image set to obtain a panoramic segmented image. This embodiment inputs the processed original image into an image processing network for sequential feature extraction, feature enhancement, and target feature redistribution, ultimately generating a panoramic segmented image corresponding to the original image. This improves the understanding of activities and events in the image scene and increases the accuracy of target object detection and recognition.
[0117] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solutions of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any restrictions on this.
[0118] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.
[0119] In addition, for technical details not described in detail in this embodiment, please refer to the image panoramic segmentation method based on feature enhancement provided in any embodiment of the present invention, which will not be repeated here.
[0120] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0121] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0122] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.
[0123] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.
Claims
1. A panoramic image segmentation method based on feature enhancement, characterized in that, include: The original image is preprocessed to obtain the processed original image; The processed original image is input into an image processing network for image feature extraction to obtain a multi-level feature image set. Image feature enhancement is performed based on the multi-level feature image set to obtain an enhanced feature image set. The enhanced feature image set is then subjected to target feature reallocation to obtain a target feature image set, specifically including: The enhanced feature image set is then merged to obtain the overall stage feature image; Target feature weights are calculated based on the enhanced feature image set to obtain stage attention weight data. The target feature image set is obtained by integrating the overall stage feature image with the stage attention weight data. The target feature image set is subjected to panoramic segmentation processing to obtain a panoramic segmented image.
2. The image panoramic segmentation method based on feature enhancement according to claim 1, characterized in that, Image feature enhancement is performed based on the multi-level feature image set to obtain an enhanced feature image set, specifically including: The feature images at different levels in the multi-level feature image set are fused according to their scale to obtain a fused feature image set. Image feature enhancement is performed on the fused feature image set to obtain the enhanced feature image set.
3. The image panoramic segmentation method based on feature enhancement according to claim 2, characterized in that, After performing image feature enhancement based on the multi-level feature image set to obtain the enhanced feature image set, the process further includes: Based on the enhanced feature image set, a convolution kernel is generated to obtain the panoramic image segmentation convolution kernel.
4. The image panoramic segmentation method based on feature enhancement according to claim 3, characterized in that, The target feature image set is subjected to panoramic segmentation processing to obtain a panoramic segmentation image, specifically including: The target feature image set is integrated and encoded to obtain the encoded target feature image; The encoded target feature image is decoded to obtain the panoramic segmentation result; The panoramic segmentation result is used to perform image post-processing to obtain the panoramic segmentation image.
5. The image panoramic segmentation method based on feature enhancement according to claim 4, characterized in that, The target feature image set is integrated and encoded to obtain the encoded target feature image, specifically including: Encoded feature extraction is performed on the target feature image set to obtain spatial coded feature images; Attention weights are redistributed based on the spatial coding feature image to obtain a unified coding feature image; Based on the panoramic segmentation convolutional kernel of the image, the unified coded feature image is segmented and predicted to obtain the coded target feature image.
6. The image panoramic segmentation method based on feature enhancement according to claim 5, characterized in that, The encoded target feature image is decoded to obtain a panoramic segmentation result, specifically including: The encoded target feature image is upsampled to obtain a sampled target feature image; The sampled target feature image is subjected to feature fusion processing to obtain the panoramic segmentation result.
7. A feature-enhanced image panoramic segmentation device, characterized in that, The feature-enhanced image panoramic segmentation device includes: Preprocessing module: preprocesses the original image to obtain the processed original image; Feature extraction module: Inputs the processed original image into an image processing network to extract image features and obtain a multi-level feature image set; Feature enhancement module: Performs image feature enhancement based on the multi-level feature image set to obtain an enhanced feature image set; Feature reassignment module: performs target feature reassignment on the enhanced feature image set to obtain a target feature image set, specifically including: The enhanced feature image set is then merged to obtain the overall stage feature image; Target feature weights are calculated based on the enhanced feature image set to obtain stage attention weight data. The target feature image set is obtained by integrating the overall stage feature image with the stage attention weight data. Panoramic segmentation module: Performs panoramic segmentation processing on the target feature image set to obtain a panoramic segmented image.
8. A feature-enhanced image panoramic segmentation device, characterized in that, The feature-enhanced image panorama segmentation device includes: a memory, a processor, and a feature-enhanced image panorama segmentation program stored in the memory and executable on the processor, wherein the feature-enhanced image panorama segmentation program is configured to implement the feature-enhanced image panorama segmentation method according to any one of claims 1 to 6.
9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it is able to implement the steps of the feature-enhanced image panoramic segmentation method according to any one of claims 1 to 6.