A low-altitude visual angle-oriented small target vehicle detection method and related device

Abstract

This application provides a method and related equipment for detecting small target vehicles from a low-altitude perspective, belonging to the fields of computer vision and intelligent transportation technology. The method includes: acquiring low-altitude images and performing multi-level feature extraction, performing feature preservation operations during the extraction process to retain high-resolution detail information; adaptively enhancing the extracted features to improve target-related feature responses; constructing a high-resolution feature response path, fusing shallow high-resolution features with deep semantic features to form enhanced features; and performing target detection based on the enhanced features. This application also optimizes feature representation through feature compression and path pruning to reduce computational complexity. Through a collaborative design of feature information preservation and reconstruction during multi-scale propagation, this application achieves a simultaneous improvement in the performance and computational efficiency of small target vehicle detection in complex low-altitude scenarios.

Description

Technical Field

[0001] This application relates to the fields of computer vision and intelligent transportation technology, and in particular to a method and related equipment for detecting small target vehicles from a low-altitude perspective. Background Technology

[0002] With the rapid development of drone technology, intelligent traffic monitoring systems based on low-altitude aerial imagery have been widely applied. In drone aerial photography scenarios, accurate detection of ground vehicles is fundamental to applications such as traffic flow monitoring, violation capture, and accident early warning.

[0003] However, vehicle detection from a low-altitude perspective faces numerous challenges. First, due to the high flight altitude of drones, target vehicles appear small in the image, typically occupying only tens to hundreds of pixels. Second, vehicles are densely distributed, often occupies space between them, and the background is complex and varied, containing roads, buildings, trees, and other elements. Under these conditions, achieving high-precision detection of small-scale targets has become a key technical challenge in this field.

[0004] In existing technologies, object detection methods are mostly based on deep learning architectures, extracting image features through multi-layer convolutional neural networks and using Feature Pyramid Networks (FPNs) or their variants to achieve multi-scale feature fusion. However, these methods suffer from the following problems when dealing with small targets at low altitudes: First, in the process of multi-layer feature extraction, as the number of network layers increases, the spatial resolution of the feature map continuously decreases. The fine-grained structural information (such as edges and textures) contained in small targets gradually decays during the propagation process, and may even be irreversibly lost. Although shallow features retain high resolution and can provide location and detail information, their semantic expressive power is weak; while deep features, although possessing strong semantic information, have severely degraded spatial structural information, making it difficult to effectively support the accurate identification of small targets.

[0005] Second, existing methods often improve detection performance by increasing network depth or introducing additional attention modules and feature enhancement structures. However, such improvements usually come at the cost of increasing model complexity, number of parameters, and computational overhead. It is difficult to ensure both detection accuracy and computational efficiency, which limits their deployment and application in resource-constrained embedded platforms or real-time detection scenarios.

[0006] Third, existing feature fusion methods mostly adopt a top-down unidirectional fusion path, which is not sufficient for the coordinated expression of shallow detail information and deep semantic information. Small targets still face the problem of insufficient feature expression during the fusion process, which can easily lead to missed detections and false detections. Summary of the Invention

[0007] The main objective of this application is to propose a method, electronic device, storage medium, and program product for detecting small target vehicles from a low-altitude perspective, in order to solve the technical problems of insufficient feature representation, weak small target discrimination ability, high false negative rate, and high computational complexity in the existing technology for detecting small target vehicles from a low-altitude perspective.

[0008] To achieve the above objectives, one aspect of this application proposes a method for detecting small target vehicles from a low-altitude perspective, the method comprising: Acquire the image to be detected from a low-altitude perspective; Multi-level feature extraction is performed on the image to be detected, and feature preservation operation is performed during the feature extraction process to preserve high-resolution detail information; Adaptive enhancement processing is applied to the extracted multi-level features to improve the response of target-related features and suppress background interference; Construct a high-resolution feature response path, and fuse shallow high-resolution features with deep semantic features to form enhanced features; Target detection is performed based on the enhanced features, and the detection results are output. The feature preservation operation includes: expanding the spatial perception range through multi-path feature extraction and fusion structure to achieve the collaborative expression of shallow detail information and deep semantic information.

[0009] In some embodiments, the multi-level feature extraction process includes: The input features are processed in parallel multipath and fused along the channel dimension to enhance the representation of local structural information.

[0010] In some embodiments, the adaptive enhancement process includes: Based on the differences in feature responses, the feature response weights at different spatial locations and channel dimensions are dynamically adjusted to enhance the feature representation of discriminative regions and suppress redundant or background interference information.

[0011] In some embodiments, constructing the high-resolution feature response path includes: Shallow high-resolution features are introduced to participate in the detection process and fused with deep semantic information across layers, thereby enhancing the characterization of fine-grained structural information while preserving semantic information.

[0012] In some embodiments, the method further includes: In the feature fusion stage, a multi-level feature collaborative expression mechanism is constructed. By establishing information transmission paths between features at different levels, high-level features can obtain detail compensation from shallow features, while shallow features are fused with deep semantic information.

[0013] In some embodiments, the method further includes a feature representation optimization step, which includes: Redundant information in the feature processing process is compressed and pruned to reduce computational complexity.

[0014] In some embodiments, the feature representation optimization step includes: A low-complexity feature transformation method is adopted for feature compression and reconstruction. By combining staged feature transformation, channel compression and local modeling, the efficient transmission of feature information is achieved. Based on the expressive contribution and computational cost of the feature paths, low-contribution feature paths are pruned, and the relevant feature interaction relationships are adjusted.

[0015] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0016] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0017] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0018] Compared with the prior art, this application has the following beneficial effects: 1) This application constructs a feature preservation mechanism to selectively preserve high-resolution detail information during the feature extraction stage, and expands the spatial perception range through multi-path feature extraction and fusion structure, effectively alleviating the attenuation problem of shallow detail information during the step-by-step propagation process, and preserving key fine-grained structural information for small target detection.

[0019] 2) This application introduces an adaptive reinforcement mechanism, which dynamically adjusts the feature response weights at different spatial locations and channel dimensions based on the differences in feature responses. This enables the network to pay more attention to key regions related to the target, while suppressing redundant or background interference information, thereby improving the discriminative power of feature representation and the robustness of the model.

[0020] 3) This application constructs a high-resolution feature response path, which integrates shallow high-resolution features with deep semantic information across layers, enabling the detection process to complete target recognition at a high spatial resolution. While maintaining semantic information, it enhances the characterization of fine-grained structural information, significantly improving the detection capability for targets at extremely small scales.

[0021] 4) This application uses a feature representation optimization mechanism to compress and trim redundant information in the feature processing process, including low-complexity feature transformation, feature compression and reconstruction, feature path trimming and other operations. While maintaining the feature representation capability, it effectively reduces the computational complexity and storage overhead, and improves the deployment efficiency of the method in resource-constrained scenarios.

[0022] 5) This application systematically and collaboratively designs mechanisms such as feature preservation, adaptive enhancement, high-resolution response and feature representation optimization, forming a complete technical solution for small target detection from a low-altitude perspective. It improves detection accuracy while taking into account computational efficiency, and has good engineering application value. Attached Figure Description

[0023] Figure 1 This is a flowchart of a small target vehicle detection method based on a low-altitude perspective provided in an embodiment of this application; Figure 2 This is a schematic diagram of the feature extraction structure for extending the sensing range in an embodiment of the present invention; Figure 3 This is a schematic diagram of the multi-path feature fusion structure in an embodiment of the present invention; Figure 4 This is a schematic diagram of the feature response adaptive adjustment mechanism in an embodiment of the present invention; Figure 5 This is a schematic diagram of the high-resolution feature response path in an embodiment of the present invention; Figure 6 This is a schematic diagram of the feature recombination processing flow in an embodiment of the present invention; Figure 7 This is a schematic diagram of the feature response path optimization process in an embodiment of the present invention; Figure 8 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0026] like Figure 1 As shown, this embodiment provides a method for detecting small target vehicles from a low-altitude perspective, specifically including the following steps: Step S1: Obtain the image to be detected from a low-altitude perspective.

[0027] Specifically, low-altitude aerial images can be acquired using cameras mounted on drones, or image data can be obtained from existing drone aerial video datasets. To improve detection performance, the acquired images can be preprocessed, including image size normalization, pixel value normalization, and data augmentation.

[0028] Step S2: Perform multi-level feature extraction on the image to be detected, and perform feature preservation operation during the feature extraction process.

[0029] In this step, a convolutional neural network is used as the backbone network for multi-level feature extraction. To preserve shallow high-resolution detail information, a feature preservation mechanism is introduced during the feature extraction process. This mechanism includes: expanding the spatial perception range through multi-path feature extraction and fusion structures, thereby achieving the collaborative expression of shallow detail information and deep semantic information.

[0030] Specifically, a multi-path feature fusion structure is constructed to integrate features extracted from different branches along the channel dimension, enhancing the richness of feature representation. By expanding the spatial perception range during feature extraction, the model can capture target information over a larger area, improving its ability to recognize small targets.

[0031] Step S3: Perform adaptive enhancement processing on the extracted multi-level features.

[0032] This step introduces an adaptive adjustment strategy based on feature response differences. By dynamically adjusting the feature responses at different spatial locations and channel dimensions, the feature representation of discriminative regions is strengthened, while redundant or background interference information is suppressed. This mechanism enables the network to focus more on key regions relevant to the target, improving the feature discrimination capability.

[0033] Step S4: Construct high-resolution feature response paths.

[0034] To address the challenge of effectively representing small targets in deep features, this step constructs a high-resolution feature response path. Shallow high-resolution features are introduced into the detection process and fused across layers with deep semantic information. This allows the feature representation process to preserve semantic information while enhancing the characterization of fine-grained structural information, thereby improving the detection capability for extremely small-scale targets.

[0035] Step S5: Perform target detection based on the enhanced features.

[0036] The enhanced features obtained through the above processing are input into the detection head, and the target category and bounding box position are output through the classification branch and regression branch to obtain the final detection result.

[0037] In a preferred embodiment, the method further includes a multi-level feature collaborative expression step. During the feature fusion stage, a multi-level feature collaborative expression mechanism is constructed. By establishing information transmission paths between features at different levels, higher-level features can obtain detail compensation from shallower layers, while shallower features can fuse with deeper semantic information, achieving consistency and completeness in feature expression.

[0038] As a further implementation method, a feature representation optimization method is also provided to improve computational efficiency. This feature representation optimization method specifically includes the following steps: This method employs a low-complexity feature transformation approach for feature compression and reconstruction. By combining staged feature transformation, channel compression, and efficient local modeling, it achieves efficient transfer of feature information. While maintaining effective feature representation, this method reduces redundant parameters involved in computation, making the feature processing more compact.

[0039] Based on the representational contribution and computational cost of feature paths, low-contribution feature paths are pruned, and related feature interaction relationships are adjusted simultaneously. For small target detection tasks from a low-altitude perspective, high-resolution feature paths express fine-grained structural information more directly, while some low-resolution paths offer limited performance gains. Pruning these paths can reduce computational complexity while preserving the responsiveness of key features.

[0040] By coordinating feature compression and path pruning, a lightweight feature processing workflow is formed, which preserves high-contribution feature paths and compresses redundant computation processes, balancing feature representation capabilities and deployment efficiency.

[0041] The solutions of the embodiments of this application will be described in detail below with reference to the accompanying drawings and specific application examples.

[0042] (1) Small target vehicle detection method for low-altitude perspective This chapter addresses the problem of insufficient feature representation in the detection of small target vehicles from a low-altitude perspective. It constructs a feature representation and reconstruction method for small targets, focusing on the feature extraction and multi-level information fusion process.

[0043] 1.1) Analysis of Feature Degradation Problem of Small Targets from Low-Altitude View In low-altitude drone aerial photography scenarios, target vehicles are typically small in scale, densely distributed, and have complex backgrounds. During multi-layer feature extraction, as features propagate through each layer, their spatial resolution continuously decreases, leading to a gradual decay or even irreversible loss of the fine-grained structural information contained in small targets.

[0044] Specifically, while shallow features offer high resolution and can preserve target edges and location information, their semantic expressive power is weak. Deep features, on the other hand, while possessing strong semantic expressive power, suffer from severe degradation of spatial structure information of small targets after multiple downsampling, making them unsuitable for effective target recognition. These issues make existing methods prone to insufficient feature representation, weak target discrimination, and high false negative rates in low-altitude vehicle detection tasks.

[0045] In addition, existing methods often improve detection performance by increasing network depth or introducing additional structural modules, but such methods often come at the cost of increasing model complexity, making it difficult to ensure both detection accuracy and computational efficiency, thus limiting their widespread application in practice.

[0046] Therefore, how to effectively preserve and reconstruct the feature information of small targets without significantly increasing computational complexity is a key technical issue for improving the performance of vehicle detection from a low-altitude perspective.

[0047] This method constructs a multi-level feature processing flow to achieve the progressive extraction and cross-layer interaction of information at different scales, thereby completing the comprehensive expression of target information. The method includes a feature extraction layer, a feature fusion layer, and a detection output layer. By progressively processing and fusing features at different levels, shallow detail information and deep semantic information are synergistically expressed, thereby improving the perception ability of small targets.

[0048] This structure achieves comprehensive modeling of information from different spatial ranges through multi-path feature extraction and fusion, thereby enhancing feature representation capabilities. During feature extraction, a multi-path structure is used to process input features in parallel and fuse them along the channel dimension, thus improving the ability to represent local structural information.

[0049] By enhancing feature responses and combining them with lightweight computational methods, a balance is achieved between feature representation capability and computational efficiency. During feature representation, key regions are enhanced, and low-computational-complexity feature processing methods are used to achieve synergistic optimization of detection performance and efficiency.

[0050] 1.2) Design of a mechanism for preserving and reconstructing small target features 1.2.1) Feature Preservation Mechanism By expanding the spatial coverage of feature extraction, comprehensive perception of information from different directions and scales can be achieved, thereby enhancing the feature representation capability in complex scenes. For example... Figure 2 As shown, by expanding the spatial perception range in the feature extraction process, the model can capture target information in a larger area, thereby enhancing its ability to identify small targets.

[0051] By constructing a multi-path feature extraction and fusion structure, full interaction of feature information is achieved, thereby improving the completeness of feature representation and discriminative ability. For example... Figure 3 As shown, by constructing a multi-path feature fusion structure, the features extracted from different branches are integrated in the channel dimension, thereby enhancing the richness of feature expression.

[0052] 1.2.2) Feature Adaptive Enhancement Mechanism In the feature representation process, an adaptive adjustment strategy based on feature response differences is introduced to enhance regions with discriminative value in the feature map while suppressing redundant or background interference information.

[0053] This mechanism dynamically adjusts feature responses at different spatial locations and channel dimensions, enabling the network to focus more on key regions relevant to the target, thereby improving feature discrimination and enhancing the model's robustness in complex scenes. Its principle is as follows: Figure 4 As shown, by adaptively adjusting the feature response, the model can focus on the target-related region, thereby improving the discriminative ability of feature representation.

[0054] 1.2.3) High-resolution feature response mechanism To address the problem that small targets are difficult to represent effectively in deep features, a high-resolution feature response path is constructed during feature processing, enabling the detection process to complete target recognition at a high spatial resolution.

[0055] See Figure 5 By introducing shallow high-resolution features into the detection process and fusing them with deep semantic information, the feature expression process can enhance the characterization of fine-grained structural information while maintaining semantic information, thereby improving the detection capability of targets at extremely small scales.

[0056] 1.3) Multi-level Feature Collaborative Representation Method In the feature fusion stage, a multi-level feature collaborative expression mechanism is constructed to achieve effective fusion of shallow detailed information and deep semantic information through cross-layer information interaction.

[0057] This method establishes information transmission paths between features at different levels, enabling higher-level features to receive detail compensation from shallower features, while shallower features can be integrated with deeper semantic information, thereby achieving consistency and completeness in feature representation. This mechanism effectively alleviates the instability of feature representation caused by scale variations and improves the model's adaptability to targets at different scales.

[0058] (2) Feature representation optimization method oriented towards computational efficiency In low-altitude small target detection tasks, the aforementioned methods enhance small target perception capabilities through multi-path feature extraction, cross-layer information reconstruction, and high-resolution feature responses to improve fine-grained feature representation. However, with the enrichment of feature representation paths, parameter redundancy, computational redundancy, and low-contribution feature paths in the feature processing process gradually emerge, affecting the deployment efficiency of the methods in resource-constrained scenarios. Building upon the aforementioned feature enhancement methods, to further improve the computational efficiency of the methods in practical applications, it is necessary to optimize and control the redundant information generated during feature processing, thereby forming an overall methodological framework that balances feature representation capabilities and computational efficiency.

[0059] To address the aforementioned issues, this chapter proposes a feature processing method optimized for computational efficiency, focusing on feature compression, path pruning, and overall lightweight construction, from the perspective of feature representation redundancy control. While preserving the expressive power of small target features as much as possible, it effectively reduces computational complexity and storage overhead, thereby improving the applicability of the method in real-time detection scenarios.

[0060] 2.1) Feature Representation Redundancy Analysis In the process of multi-level feature representation, while the expressive power of small targets is enhanced with the increase of feature extraction paths, fusion paths, and response paths, the feature processing chain becomes longer, with information overlap between some feature branches and repetitive local computation processes. Especially in low-altitude small target detection tasks, high-resolution feature paths contribute more significantly to the detection results, while some low-contribution paths introduce significant computational overhead without improving the overall detection performance.

[0061] Therefore, it is necessary to analyze the redundant structures in the feature representation process and optimize the feature processing flow from the perspectives of feature contribution differences, path importance, and computational cost distribution, so as to achieve a balance between detection performance and computational efficiency.

[0062] 2.2) Computational complexity optimization mechanism Based on the above analysis, this chapter proposes a computational complexity optimization mechanism by starting with the relationship between feature representation and computational complexity. This mechanism mainly consists of three parts: first, feature compression and reconstruction through low-complexity feature transformation; second, pruning and optimizing feature paths based on differences in feature contribution; and third, building a lightweight feature processing flow based on the above to reduce overall computational overhead and improve actual deployment efficiency.

[0063] 2.2.1) Feature Compression and Reconstruction Methods In feature processing, while some standard feature transformation operations possess strong expressive power, they also introduce a high number of parameters and computational overhead. To reduce this computational cost, see [reference needed]. Figure 6 This paper proposes a low-complexity feature compression and reconstruction method, which achieves efficient transfer of feature information by combining staged feature transformation, channel compression and efficient local modeling.

[0064] This method reduces redundant parameters in the calculation while maintaining effective feature representation, making the feature processing more compact. Compared with directly reducing the number of feature layers, this method does not simply sacrifice feature expressive power, but maintains the ability to represent the target features while reducing computational complexity through a more efficient feature recombination method.

[0065] 2.2.2) Feature path pruning mechanism In the process of multi-level feature representation, different feature paths contribute differently to the detection results. For small target detection tasks from a low-altitude perspective, high-resolution feature paths are more direct in expressing fine-grained structural information, while some low-resolution paths, although introducing additional computational overhead, have relatively limited gains on small target detection performance.

[0066] Based on this, this paper proposes a feature path pruning mechanism, such as... Figure 7 The mechanism described above prunes low-contribution paths based on their expressive contribution and computational cost, and simultaneously adjusts related feature interactions to reduce redundant computations. In this way, the feature processing flow is simplified and computational complexity is reduced while preserving the responsiveness of key features.

[0067] 2.2.3) Lightweight Model Construction Method Building upon feature compression and path optimization, a lightweight feature processing workflow is further constructed. This workflow coordinates the feature extraction, feature fusion, and detection response processes as a whole, preserving high-contribution feature paths and compressing redundant computation processes, thus forming a lightweight detection method that balances feature representation capabilities and deployment efficiency.

[0068] Unlike methods that compress only local structures, this method starts from the overall feature processing chain and coordinates the computational overhead at different stages. This allows the constructed method to maintain good small target detection performance even in resource-constrained scenarios, making it highly valuable for engineering applications.

[0069] In summary, this embodiment proposes a method for detecting small target vehicles from a low-altitude perspective and its feature reconstruction mechanism. Addressing the issue that small-scale target vehicles in UAV aerial photography scenarios are prone to gradual attenuation or even irreversible loss of detail information during multi-layer feature extraction and transmission, resulting in insufficient feature representation capabilities for small targets, this invention constructs a multi-level feature processing mechanism based on feature preservation and cross-layer reconstruction. By selectively preserving high-resolution detail information during the feature extraction stage and introducing cross-layer information compensation and reconstruction strategies during feature fusion, the collaborative expression of shallow detail information and deep semantic information is achieved, thereby improving the feature recognition capability of small targets.

[0070] Building upon this foundation, a high-resolution feature response path for small-scale targets is further constructed, and a redundant feature suppression strategy is combined to compress and optimize feature representation, reducing computational complexity while maintaining detection performance. This method, through the coordinated design of feature information preservation and reconstruction during multi-scale propagation, achieves simultaneous improvement in the detection performance and computational efficiency of small-target vehicles in complex low-altitude scenarios, demonstrating significant engineering application value.

[0071] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0072] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0073] Please see Figure 8 , Figure 8 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 801 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 802 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 802 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 802 and is called and executed by the processor 801 using the methods described in the embodiments of this application. The 803 input / output interface is used to implement information input and output. The communication interface 804 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 805 transmits information between various components of the device (e.g., processor 801, memory 802, input / output interface 803, and communication interface 804); The processor 801, memory 802, input / output interface 803, and communication interface 804 are connected to each other within the device via bus 805.

[0074] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0075] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0076] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0077] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0078] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented in the embodiments of this program product are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments. The executable computer program code or "code" used to perform the various embodiments can be written in high-level programming languages ​​such as C, C++, Python, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0079] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0080] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0081] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0082] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0083] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application 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 this application 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 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.

[0084] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0085] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0086] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0087] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0088] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0089] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for detecting small target vehicles from a low-altitude perspective, characterized in that, The method includes the following steps: Acquire the image to be detected from a low-altitude perspective; Multi-level feature extraction is performed on the image to be detected, and feature preservation operation is performed during the feature extraction process to preserve high-resolution detail information; Adaptive enhancement processing is applied to the extracted multi-level features to improve the response of target-related features and suppress background interference; Construct a high-resolution feature response path, and fuse shallow high-resolution features with deep semantic features to form enhanced features; Target detection is performed based on the enhanced features, and the detection results are output. The feature preservation operation includes: expanding the spatial perception range through multi-path feature extraction and fusion structure to achieve the collaborative expression of shallow detail information and deep semantic information.

2. The method according to claim 1, characterized in that, The multi-level feature extraction process includes: The input features are processed in parallel multipath and fused along the channel dimension to enhance the representation of local structural information.

3. The method according to claim 1, characterized in that, The adaptive enhancement process includes: Based on the differences in feature responses, the feature response weights at different spatial locations and channel dimensions are dynamically adjusted to enhance the feature representation of discriminative regions and suppress redundant or background interference information.

4. The method according to claim 1, characterized in that, The construction of the high-resolution feature response path includes: Shallow high-resolution features are introduced to participate in the detection process and fused with deep semantic information across layers, thereby enhancing the characterization of fine-grained structural information while preserving semantic information.

5. The method according to claim 1, characterized in that, The method further includes: In the feature fusion stage, a multi-level feature collaborative expression mechanism is constructed. By establishing information transmission paths between features at different levels, high-level features can obtain detail compensation from shallow features, while shallow features are fused with deep semantic information.

6. The method according to claim 1, characterized in that, The method further includes a feature representation optimization step, which includes: Redundant information in the feature processing process is compressed and pruned to reduce computational complexity.

7. The method according to claim 6, characterized in that, The feature representation optimization step includes: A low-complexity feature transformation method is adopted for feature compression and reconstruction. By combining staged feature transformation, channel compression and local modeling, the efficient transmission of feature information is achieved. Based on the expressive contribution and computational cost of the feature paths, low-contribution feature paths are pruned, and the relevant feature interaction relationships are adjusted.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images