Unmanned aerial vehicle dynamic detection system based on image recognition and efficient communication

Abstract

The application provides a UAV dynamic detection system based on image recognition and efficient communication, solves the problem of low image recognition accuracy by processing and recognizing images collected by a UAV detection terminal, provides a high-quality data source for subsequent image classification and trajectory calculation, guarantees the subsequent trajectory accuracy and recognition reliability, classifies the UAV images based on the proportion of effective information of the UAV to obtain multi-level image data, separates the core detection data and redundant data, reduces the transmission load from the data source, solves the problem of insufficient communication efficiency, determines the encoding mode and channel allocation of the multi-level image data based on the real-time state of the communication channel, and transmits the multi-level image data, differentiates the channel allocation of the multi-level image data, allocates high-bandwidth and low-packet-loss-rate high-quality channels to high-effective core data, allocates ordinary channels to low-effective data, optimally utilizes the channel resources, and significantly improves the transmission efficiency.

Description

Technical Field

[0001] This invention relates to the fields of unmanned aerial vehicle (UAV) technology, image recognition technology, and wireless communication technology, and particularly to a UAV dynamic detection system based on image recognition and efficient communication. Background Technology

[0002] In recent years, the adoption of consumer and industrial drones has accelerated. In scenarios such as security patrols, border control, and airspace security, the demand for dynamic detection of illegally intruding drones is increasing. Existing drone detection systems generally suffer from two major technological flaws, which have become key bottlenecks restricting the practicality of these systems: Low image recognition accuracy and poor adaptability to complex scenes: Existing detection drones mostly use general target detection models, which are not optimized for low-altitude, small-sized drone targets. When shooting from high altitudes at long distances, intruding drones occupy a very small area in the frame, making it easy to miss small targets; drone flight vibrations and airflow disturbances can cause motion blur in the image; background interference such as trees, buildings, and clouds can easily lead to misidentification; at the same time, single-scale feature extraction cannot take into account both large targets in the foreground and small targets in the background, resulting in insufficient overall recognition accuracy and high false alarm and missed alarm rates; Wireless communication is inefficient, making it difficult to balance real-time performance and reliability. On the one hand, high-definition detection images contain massive amounts of data, and direct transmission would consume a large amount of wireless bandwidth, resulting in high transmission latency and severe stuttering, which cannot meet the requirements for real-time tracking of dynamic targets. On the other hand, existing communication systems mostly use fixed bandwidth and fixed channels for transmission, without distinguishing the priority of detection data. Even if invalid background information is identified, it is still transmitted at full speed, resulting in a waste of bandwidth resources. In long-distance, low-altitude environments with multiple interferences, data packet loss and link interruptions are also prone to occur, further reducing communication efficiency and system stability. Summary of the Invention

[0003] This invention provides a dynamic detection system for unmanned aerial vehicles (UAVs) based on image recognition and efficient communication, in order to solve the problems mentioned in the background art.

[0004] A dynamic detection system for unmanned aerial vehicles (UAVs) based on image recognition and efficient communication includes: The drone identification module is used to process and identify images collected by the drone detection terminal to obtain drone images; The image grading module is used to grade UAV images based on valid UAV information to obtain multi-level image data. The data transmission module is used to determine the encoding method and channel allocation for multi-level image data based on the real-time status of the communication channel, and to transmit the multi-level image data. The trajectory display module is used to determine and display the dynamic trajectory of the UAV based on the received multi-level image data.

[0005] Preferably, the drone identification module includes: The repair unit is used to repair the acquired images based on the attitude data of the UAV synchronously collected by the UAV detection terminal, and to obtain the UAV repaired image based on the attitude data. The multi-scale feature extraction unit is used to extract features from UAV restored images at micro, small, medium and large scales to obtain micro-scale features, small-scale features, medium-scale features and large-scale features; The weight determination unit is used to unify the dimensions of micro-scale features, small-scale features, medium-scale features and large-scale features to obtain feature vectors at each scale. It calculates the cosine similarity between the feature vectors at each scale and the standard feature template of the UAV, determines the initial weights based on the cosine similarity, and determines the attention weights of the feature vectors at each scale based on the pre-designed spatial weight distribution map. It also determines the spatial attention weights at each scale based on the initial weights and attention weights. The weight determination unit is also used to obtain the feature vectors of 5 consecutive frames at each scale, calculate the feature similarity between the last frame and the previous 4 consecutive frames, determine the temporal attention weight of the current frame based on the similarity, and use the product of the spatial attention weight and the temporal attention weight as the comprehensive weight of the feature vector at each scale. The feature fusion unit is used to perform weighted fusion of feature vectors at all scales based on comprehensive weights to obtain initial fused features. It also analyzes the proportion of effective feature quantity in the feature vectors at each scale, retains feature vectors with an effective feature quantity proportion greater than 0.5, and suppresses feature vectors with an effective feature quantity proportion not greater than 0.5 to obtain effective feature vectors at all scales. It obtains the difference between the feature vector and the effective feature vector at each scale, and performs weighted summation of the difference based on comprehensive weights to obtain residual fused features. Finally, it adds the initial fused features and the residual fused features across all dimensions to obtain the target fused feature vector. The image determination unit is used to obtain UAV images based on target fusion feature vector integration.

[0006] Preferably, the repair unit includes: The acquisition unit is used to acquire the angular velocity, acceleration, and heading angle of the UAV based on the attitude sensor set on the UAV detection terminal as attitude data, and to perform spatiotemporal alignment of the attitude data and the UAV image to obtain aligned attitude data; The judgment unit is used to determine the degree of blur in the UAV image based on the jitter parameters of the aligned attitude data and the mapping relationship between jitter and image blur. The image restoration unit is used to determine restoration parameters based on jitter parameters when the blur level is greater than a preset level, and to restore the acquired image based on the restoration parameters to obtain a UAV restored image.

[0007] Preferably, the multi-scale feature extraction unit includes: The microscale feature extraction unit is used to obtain a first image from the UAV repair image with a detection distance within a first preset distance range, and extract the contour features and overall rotor features of the first image as microscale features; The small-scale feature extraction unit is used to obtain a second image from the UAV repair image with a detection distance within a second preset distance range, and extract the fuselage detail features and rotor rotation features of the second image as small-scale features; The mesoscale feature extraction unit is used to obtain a third image from the UAV repair image with a detection distance within a third preset distance range, and extract the fuselage identification features, engine features and rotor blade features of the third image as mesoscale features; The large-scale feature extraction unit is used to obtain a fourth image from the UAV repair image with a detection distance within a fourth preset distance range, and extract the fuselage structure features, rotor speed features, and landing gear features of the fourth image as large-scale features.

[0008] Preferably, the image grading module includes: The setting unit is used to set effective information features and their corresponding priorities based on the application of historical flight images of UAVs. The level unit is used to compare the UAV image with the effective information features. When the similarity is greater than the preset similarity, the priority of the effective information features is determined as the priority of the UAV image. The data determination unit is used to obtain multi-level image data based on the priority of UAV images.

[0009] Preferably, the data transmission module includes: The encoding setting unit is used to design a multi-layer encoding structure consisting of a base layer, a reinforcement layer, and an enhancement layer. High-priority data is encoded using the base layer, reinforcement layer, and enhancement layer; medium-priority data is encoded using the base layer and reinforcement layer; and low-priority data is encoded using the base layer, thus establishing the encoding rules. The compression ratio setting unit is used to set the coding compression ratio from low to high for the multi-layer coding structure according to the channel quality from high to low, establish the compression rules, and establish the basic coding mechanism based on the coding rules and compression rules. The channel allocation unit is used to determine the real-time quality level of the communication channel based on the real-time status of the communication channel, and to allocate communication channels for multi-level image data based on the real-time quality level. The transmission unit is used to transmit multi-level image data after encoding and compressing it according to the allocation of communication channels for multi-level image data and the basic coding mechanism.

[0010] Preferably, the channel allocation unit includes: The level determination unit is used to obtain the channel parameters of the real-time status of the communication channel and determine the real-time quality level of the communication channel based on the correspondence between the quality level and the channel parameters. The allocation setting unit is used to set the channel selection priority of high-priority data from high to low as high-quality channel, good channel, and normal channel; the channel selection priority of medium-priority data from high to low as good channel, high-quality channel, and normal channel; and the channel selection priority of low-priority data from high to low as normal channel, good channel, and high-quality channel. The initial communication channel is allocated to multi-level image data in the above manner. The revenue determination unit is used to establish the target revenue function based on the average utilization, average transmission delay, and average packet loss rate of communication channels at different quality levels. The initial allocation unit is used to determine the revenue values ​​of high-priority data, medium-priority data, and low-priority data based on the target revenue function. When all revenue values ​​are greater than the preset revenue value, the initial communication channel remains unchanged; otherwise, the initial communication channel is allocated and adjusted. The real-time adjustment unit is used to allocate and adjust the initial communication channel when it is necessary. The first priority principle is that the overall benefit value is not less than the minimum threshold. The second priority principle is to ensure the stable transmission of high-priority data, medium-priority data and low-priority data. The unit allocates communication channels for multi-level image data according to the allocation and adjustment results.

[0011] Preferably, the target return function in the return determination unit Specifically as follows: in, Indicates channel utilization weights, This represents the average utilization rate of all communication channels at the given quality level. Indicates the channel transmission delay weight. This represents the average transmission delay of all communication channels at the given quality level. Indicates the channel packet loss rate weight. This represents the average packet loss rate across all communication channels at the given quality level.

[0012] Preferably, the trajectory display module includes: The receiving unit is used to receive multi-level image data transmitted by the data transmission module; The extraction unit is used to extract the target coordinates of the UAV from multi-level image data, prioritizing high-priority data, medium-priority data, and low-priority data as auxiliary data. The trajectory generation unit is used to fit a dynamic trajectory based on the target coordinates of the UAV and then visualize it.

[0013] Compared with the prior art, the present invention has achieved the following beneficial effects: By processing and recognizing images collected by UAV detection terminals, the problem of low image recognition accuracy is solved, significantly reducing the probability of false detection and missed detection of targets. This provides a high-quality data source for subsequent image grading and trajectory calculation, ensuring the accuracy and reliability of subsequent trajectory recognition. By grading UAV images based on the proportion of effective UAV information, multi-level image data is obtained, achieving the separation of core detection data and redundant data. This breaks the traditional full-volume high-definition transmission mode, reducing the transmission load from the data source and solving the problem of insufficient communication efficiency. Based on the real-time status of the communication channel, the encoding method and channel allocation for multi-level image data are determined, and multi-level image data is transmitted. Channels are allocated differently for multi-level image data. High-effective-level core data is allocated to high-bandwidth, low-packet-loss-rate premium channels, while low-effective-level data is allocated to ordinary channels, achieving optimal utilization of channel resources, significantly improving transmission efficiency and reducing transmission latency. The dynamic trajectory of the UAV is determined and displayed based on the received multi-level image data. The trajectory is calculated based on the graded transmission of image data, with high-priority target data arriving and being parsed first, ensuring strong real-time, lag-free, and uninterrupted dynamic trajectory updates, meeting the real-time monitoring requirements of dynamic detection.

[0014] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural diagram of the UAV dynamic detection system based on image recognition and efficient communication in an embodiment of the present invention; Figure 2 This is a structural diagram of the image grading module in an embodiment of the present invention; Figure 3 This is a structural diagram of the data transmission module in an embodiment of the present invention. Detailed Implementation

[0017] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0018] Example 1: This embodiment of the invention provides a dynamic detection system for unmanned aerial vehicles (UAVs) based on image recognition and efficient communication, such as... Figure 1 As shown, it includes: The drone identification module is used to process and identify images collected by the drone detection terminal to obtain drone images; The image grading module is used to grade UAV images based on valid UAV information to obtain multi-level image data. The data transmission module is used to determine the encoding method and channel allocation for multi-level image data based on the real-time status of the communication channel, and to transmit the multi-level image data. The trajectory display module is used to determine and display the dynamic trajectory of the UAV based on the received multi-level image data.

[0019] In this embodiment, the images collected by the UAV detection terminal are infrared thermal imaging images.

[0020] In this embodiment, the effective information ratio of the UAV refers to the proportion of effective detection information such as the pixel ratio, feature integrity, and target clarity of the UAV target area in a single frame image, in the total information of the entire frame image.

[0021] In this embodiment, the drone images are classified into multi-level image data by dividing the identified images into multi-level image data with different priorities.

[0022] In this embodiment, the real-time status of the communication channel includes real-time operating parameters such as channel signal-to-noise ratio, transmission bandwidth, packet loss rate, transmission delay, and electromagnetic interference intensity at the current moment.

[0023] In this embodiment, the dynamic trajectory of the UAV is a continuous motion trajectory composed of the UAV's real-time position, direction of movement, and flight speed, calculated from multiple consecutive frames of UAV image data.

[0024] The beneficial effects of the above design scheme are as follows: By processing and recognizing the images collected by the UAV detection terminal, the problem of low image recognition accuracy is solved, the probability of false detection and missed detection of targets is greatly reduced, and a high-quality data source is provided for subsequent image classification and trajectory calculation, ensuring the accuracy and reliability of subsequent trajectory recognition. By classifying UAV images based on the proportion of effective information of the UAV, multi-level image data is obtained, realizing the separation of core detection data and redundant data, breaking the traditional full-volume high-definition transmission mode, reducing the transmission load from the data source, and solving the problem of insufficient communication efficiency. By determining the encoding method and channel allocation of multi-level image data based on the real-time status of the communication channel, multi-level image data is transmitted. Channels are allocated differently for multi-level image data. High-effective-level core data is allocated to high-bandwidth, low-packet-loss-rate high-quality channels, and low-effective-level data is allocated to ordinary channels, realizing the optimal utilization of channel resources, significantly improving transmission efficiency and reducing transmission latency. The dynamic trajectory of the UAV is determined and displayed based on the received multi-level image data. The trajectory is calculated based on the image data transmitted in a graded manner, and high-priority target data arrives and is parsed first, ensuring that the dynamic trajectory update is real-time, without lag or interruption, and meeting the real-time monitoring requirements of dynamic detection.

[0025] Example 2: Based on Example 1, this embodiment of the invention provides a UAV dynamic detection system based on image recognition and efficient communication. The UAV recognition module includes: The repair unit is used to repair the acquired images based on the attitude data of the UAV synchronously collected by the UAV detection terminal, and to obtain the UAV repaired image based on the attitude data. The multi-scale feature extraction unit is used to extract features from UAV restored images at micro, small, medium and large scales to obtain micro-scale features, small-scale features, medium-scale features and large-scale features; The weight determination unit is used to unify the dimensions of micro-scale features, small-scale features, medium-scale features and large-scale features to obtain feature vectors at each scale. It calculates the cosine similarity between the feature vectors at each scale and the standard feature template of the UAV, determines the initial weights based on the cosine similarity, and determines the attention weights of the feature vectors at each scale based on the pre-designed spatial weight distribution map. It also determines the spatial attention weights at each scale based on the initial weights and attention weights. The weight determination unit is also used to obtain the feature vectors of 5 consecutive frames at each scale, calculate the feature similarity between the last frame and the previous 4 consecutive frames, determine the temporal attention weight of the current frame based on the similarity, and use the product of the spatial attention weight and the temporal attention weight as the comprehensive weight of the feature vector at each scale. The feature fusion unit is used to perform weighted fusion of feature vectors at all scales based on comprehensive weights to obtain initial fused features. It also analyzes the proportion of effective feature quantity in the feature vectors at each scale, retains feature vectors with an effective feature quantity proportion greater than 0.5, and suppresses feature vectors with an effective feature quantity proportion not greater than 0.5 to obtain effective feature vectors at all scales. It obtains the difference between the feature vector and the effective feature vector at each scale, and performs weighted summation of the difference based on comprehensive weights to obtain residual fused features. Finally, it adds the initial fused features and the residual fused features across all dimensions to obtain the target fused feature vector. The image determination unit is used to obtain UAV images based on target fusion feature vector integration.

[0026] In this embodiment, the initial fusion features are the sum of the feature values ​​of micro-scale features, small-scale features, medium-scale features, and large-scale features after multiplying them by their corresponding comprehensive weights.

[0027] In this embodiment, the dimensional unification of micro-scale features, small-scale features, medium-scale features, and large-scale features specifically involves converting features of the four scales into 384-dimensional feature vectors. Specifically, feature vectors with dimensions greater than 384 are compressed to 384 dimensions through principal component analysis, while feature vectors with dimensions less than 384 are zero-padded to 384 dimensions.

[0028] In this embodiment, the residual fusion feature is the sum of the differences between the feature vectors of microscale features, small-scale features, medium-scale features, and large-scale features and the effective feature vectors, respectively, after multiplying them by the corresponding comprehensive weights.

[0029] In this embodiment, the spatial weight distribution map specifically assigns high weight to key areas and low weight to irrelevant areas.

[0030] In this embodiment, the standard feature template of the UAV is pre-trained based on feature data of various types and models of UAVs.

[0031] In this embodiment, the effective feature quantity ratio is the ratio of the effective feature quantity to the total feature quantity in the feature vector, and the effective feature quantity is the number of features that have a matching degree greater than a preset matching degree with the standard feature template of the UAV.

[0032] The beneficial effects of the above design scheme are as follows: By synchronously collecting the attitude data of the UAV based on the UAV detection terminal, and repairing the collected images based on the attitude data, a UAV restored image is obtained, solving the motion blur problem and improving the basic image quality. By extracting features from the UAV restored image at micro, small, medium, and large scales, micro-scale features, small-scale features, medium-scale features, and large-scale features are obtained, comprehensively capturing the feature details of UAVs at different distances. The feature extraction at different scales is highly targeted, with each scale feature having its own emphasis and no redundancy. Micro-scale features focus on contours, while large-scale features focus on details, improving recognition accuracy. The features are unified in dimensionality to obtain feature vectors for each scale, avoiding fusion distortion caused by dimensional differences and balancing fusion efficiency and feature integrity. Initial weights are determined based on the cosine similarity between the feature vectors at each scale and the standard feature template of the UAV. The higher the similarity, the greater the weight, which prioritizes strengthening scale features with high matching degree with the core features of the UAV, avoiding irrelevant features from dominating the fusion result and improving the rationality of weight allocation. By combining a pre-designed spatial weight distribution map, the attention weights of the feature vectors at each scale are adjusted to accurately strengthen the feature weights of the UAV target area and suppress the invalid feature weights of the background area, further improving feature recognition. By obtaining the feature vectors of 5 consecutive frames, the current frame and the previous 4 frames are compared. Feature similarity is used to determine temporal attention weights, which effectively utilizes the temporal dimension features of consecutive frames, solving the recognition bias caused by incomplete features in a single frame image. This makes weight allocation more dynamic and complete. By multiplying spatial attention weights and temporal attention weights to obtain a comprehensive weight, both the importance of features at each scale in a single frame image and the correlation of features in consecutive frames are considered, avoiding the limitations of a single spatial or temporal weight. This ensures that subsequent feature fusion can focus on the core features of the UAV, improving recognition accuracy and stability. By weighting and fusing feature vectors at all scales using the comprehensive weight, core features with high weights can be retained first, while redundant features with low weights can be suppressed, improving the discriminative power of the initial fused features. By calculating the difference between the original feature vector and the effective feature vector at each scale, and then weighting and summing the differences based on the comprehensive weights to obtain the residual fusion feature, and then adding it to the initial fusion feature across all dimensions, the system can effectively compensate for the small number of key details lost during the effective feature selection process, avoid fusion feature distortion, and ensure that the target fusion feature vector can comprehensively and accurately represent the UAV target. The entire fusion process does not require complex calculations. The weight allocation and residual calculation are based on the existing feature and weight data mentioned above, resulting in low computational load. This enables rapid fusion of multi-scale features, adapts to the real-time recognition requirements of UAV dynamic detection, avoids target tracking lag caused by excessive fusion time, and provides high-quality UAV image data for subsequent image grading and efficient communication.

[0033] Example 3: Based on Example 2, this embodiment of the invention provides a UAV dynamic detection system based on image recognition and efficient communication. The repair unit includes: The acquisition unit is used to acquire the angular velocity, acceleration, and heading angle of the UAV based on the attitude sensor set on the UAV detection terminal as attitude data, and to perform spatiotemporal alignment of the attitude data and the UAV image to obtain aligned attitude data; The judgment unit is used to determine the degree of blur in the UAV image based on the jitter parameters of the aligned attitude data and the mapping relationship between jitter and image blur. The image restoration unit is used to determine restoration parameters based on jitter parameters when the blur level is greater than a preset level, and to restore the acquired image based on the restoration parameters to obtain a UAV restored image.

[0034] In this embodiment, the jitter parameters include jitter direction, amplitude, and frequency.

[0035] In this embodiment, the repair parameters are determined based on the pre-designed correspondence between the jitter parameters and the repair parameters. This correspondence is determined based on the repair status of historical images. The jitter direction determines the direction of blur offset, the jitter amplitude determines the severity of blur, and the jitter frequency determines the density of blur.

[0036] The beneficial effects of the above design scheme are as follows: By collecting three core attitude data of the UAV—angular velocity, acceleration, and heading angle—it covers the key parameters of UAV flight jitter, comprehensively characterizes changes in UAV flight attitude, accurately captures the core features of jitter, and provides high-quality data support for subsequent fuzzy judgment and repair parameter determination. By spatiotemporally aligning the collected attitude data with UAV images, it ensures that each frame corresponds to unique attitude data, making subsequent fuzzy judgment and image repair based on jitter parameters more targeted. By using jitter parameters based on aligned attitude data, combined with the mapping relationship between jitter and image blur, it can determine the flight attitude of the UAV. The system comprehensively and accurately reflects the blur level of human-machine images. It can quickly retrieve the corresponding blur level directly through aligned jitter parameters, eliminating the need for complex blur detection algorithms, reducing the computational load of the judgment unit, improving the efficiency of blur judgment, and adapting to the real-time requirements of UAV dynamic detection. At the same time, it ensures a unified judgment standard and avoids the subjective bias of human judgment. When the blur level is greater than the preset level, the system determines the repair parameters based on the jitter parameters, ensuring that the repair parameters accurately correspond to the jitter state. This ensures that the repaired UAV image can clearly retain core features, providing a high-purity data source for subsequent multi-scale feature extraction.

[0037] Example 4: Based on Example 2, this embodiment of the invention provides a UAV dynamic detection system based on image recognition and efficient communication. The multi-scale feature extraction unit includes: The microscale feature extraction unit is used to obtain a first image from the UAV repair image with a detection distance within a first preset distance range, and extract the contour features and overall rotor features of the first image as microscale features; The small-scale feature extraction unit is used to obtain a second image from the UAV repair image with a detection distance within a second preset distance range, and extract the fuselage detail features and rotor rotation features of the second image as small-scale features; The mesoscale feature extraction unit is used to obtain a third image from the UAV repair image with a detection distance within a third preset distance range, and extract the fuselage identification features, engine features and rotor blade features of the third image as mesoscale features; The large-scale feature extraction unit is used to obtain a fourth image from the UAV repair image with a detection distance within a fourth preset distance range, and extract the fuselage structure features, rotor speed features, and landing gear features of the fourth image as large-scale features.

[0038] In this embodiment, the first preset distance is 800-1000m, the second preset distance is 500-800m, the third preset distance is 200-500m, and the fourth preset distance is 0-200m.

[0039] The beneficial effects of the above design scheme are as follows: By accurately adapting to the first preset distance range to obtain micro-scale features, the overall outline and rotor distribution of the drone are captured, which avoids the waste of computing power caused by ineffective detail extraction in long-distance scenes and can accurately distinguish the drone from distant background interference. By accurately adapting to the second preset distance range to obtain small-scale features, the body detail features and rotor rotation features are extracted in a targeted manner, which is consistent with the characteristics of the drone outline being clear and details being distinguishable at this distance but not reaching the close-up standard, thus greatly improving the feature discrimination. By accurately adapting to the third preset distance range to obtain medium-scale features, more comprehensive core features are provided, which provides strong support for subsequent accurate recognition, while avoiding the high computing power consumption of large-scale detail extraction. By accurately adapting to the fourth preset distance range to obtain large-scale features, which is consistent with the characteristics of clear details and complete features of close-range drones, laying the foundation for the accurate generation of target fusion feature vectors.

[0040] Example 5: Based on Example 1, this embodiment of the invention provides a UAV dynamic detection system based on image recognition and efficient communication, such as... Figure 2 As shown, the image grading module includes: The setting unit is used to set effective information features and their corresponding priorities based on the application of historical flight images of UAVs. The level unit is used to compare the UAV image with the effective information features. When the similarity is greater than the preset similarity, the priority of the effective information features is determined as the priority of the UAV image. The data determination unit is used to obtain multi-level image data based on the priority of UAV images.

[0041] In this embodiment, effective information features, such as UAV target information, are given high priority, background environment information is given medium priority, and other information is given low priority.

[0042] The beneficial effects of the above design scheme are as follows: Based on the application of historical flight images of UAVs, and combined with scenario-based data such as historical recognition accuracy, data transmission efficiency, and actual application needs, effective information features and corresponding priorities are set. This ensures that the set features are core information with high value in actual detection, and the priority division is in line with actual application scenarios. By comparing the similarity between UAV images and the effective information features set by the setting unit, when the similarity meets the standard, the priority of the corresponding effective information feature is directly determined as the priority of the UAV image. High-value effective information corresponds to high priority, and low-value information corresponds to low priority. This provides a precise basis for the differentiated scheduling of subsequent data transmission and avoids core data being occupied by low-priority data.

[0043] Example 6: Based on Example 1, this embodiment of the invention provides a UAV dynamic detection system based on image recognition and efficient communication, such as... Figure 3 As shown, the data transmission module includes: The encoding setting unit is used to design a multi-layer encoding structure consisting of a base layer, a reinforcement layer, and an enhancement layer. High-priority data is encoded using the base layer, reinforcement layer, and enhancement layer; medium-priority data is encoded using the base layer and reinforcement layer; and low-priority data is encoded using the base layer, thus establishing the encoding rules. The compression ratio setting unit is used to set the coding compression ratio from low to high for the multi-layer coding structure according to the channel quality from high to low, establish the compression rules, and establish the basic coding mechanism based on the coding rules and compression rules. The channel allocation unit is used to determine the real-time quality level of the communication channel based on the real-time status of the communication channel, and to allocate communication channels for multi-level image data based on the real-time quality level. The transmission unit is used to transmit multi-level image data after encoding and compressing it according to the allocation of communication channels for multi-level image data and the basic coding mechanism.

[0044] In this embodiment, according to the channel quality from high to low, the compression ratios of the multi-layer coding structure are set from low to high as follows: in a high-quality channel, the compression ratios of the base layer, reinforcement layer, and enhancement layer are 1:2 (compressed data size: original data size), 1:3, and 2:5, respectively; in a good channel, the compression ratios of the base layer, reinforcement layer, and enhancement layer are 1:3, 1:4, and 2:6, respectively; and in a normal channel, the compression ratios of the base layer, reinforcement layer, and enhancement layer are 1:4, 1:5, and 2:7, respectively.

[0045] In this embodiment, the channel quality is ranked from high to low as high quality channel, good channel, and normal channel.

[0046] In this embodiment, the real-time status of the communication channel includes electromagnetic interference intensity, transmission bandwidth, packet loss rate, and channel signal-to-noise ratio, etc. The real-time quality level is determined by pre-designing the correspondence between the quality level and the channel parameters.

[0047] The beneficial effects of the above design scheme are as follows: By establishing differentiated coding rules based on the priority differences of multi-level image data, high-priority data is encoded using the base layer, enhancement layer, and reinforcement layer to ensure the complete preservation of core and detailed features; medium-priority data is encoded using the base layer and reinforcement layer to balance feature integrity and transmission efficiency; low-priority data is encoded only using the base layer to eliminate redundant features and reduce coding resource consumption. This rule avoids the problems of wasting resources with full-scale encoding and losing core features with simplified encoding, achieving optimal allocation of coding resources. By setting compression ratios from low to high according to the quality levels of high-quality channels to good channels to ordinary channels, the compression ratio is set to match the transmission capabilities of different channels. High-quality channels have stable transmission and sufficient bandwidth, so a low compression ratio is used to ensure the feature integrity of the encoded data; good channels have relatively stable transmission and moderate bandwidth, so a medium compression ratio is used to balance feature integrity and transmission efficiency; ordinary channels have unstable transmission and limited bandwidth, so a high compression ratio is used to prioritize the protection of core features. This system improves transmission efficiency, reduces bandwidth usage, and addresses the poor channel adaptability of conventional fixed compression ratios. It ensures a high degree of match between compression rules and channel quality by establishing a fundamental coding mechanism that clarifies the coding levels of each data level and the corresponding channel compression ratio standards. This makes coding and compression operations systematic and avoids mismatches between coding structure and compression ratio caused by a disconnect between coding and compression. The hierarchical adaptation design of the compression ratio balances the integrity of core features with bandwidth utilization, ultimately guaranteeing stable and efficient transmission in complex channel environments. Real-time channel quality grading provides a reliable basis for accurate allocation. Based on the real-time channel quality level and the priority of multi-level image data, a principle is established to prioritize high-priority data for high-quality channels, ensuring the real-time and complete transmission of core feature data. By strictly adhering to the communication channels allocated by the channel allocation unit and combining the fundamental coding mechanism established by the compression ratio setting unit, image data at each level is encoded and compressed before transmission, ensuring a standardized transmission process, accurate data transmission, and reducing the risk of data distortion and packet loss during transmission.

[0048] Example 7: Based on Example 6, this embodiment of the invention provides a UAV dynamic detection system based on image recognition and efficient communication, wherein the channel allocation unit includes: The level determination unit is used to obtain the channel parameters of the real-time status of the communication channel and determine the real-time quality level of the communication channel based on the correspondence between the quality level and the channel parameters. The allocation setting unit is used to set the channel selection priority of high-priority data from high to low as high-quality channel, good channel, and normal channel; the channel selection priority of medium-priority data from high to low as good channel, high-quality channel, and normal channel; and the channel selection priority of low-priority data from high to low as normal channel, good channel, and high-quality channel. The initial communication channel is allocated to multi-level image data in the above manner. The revenue determination unit is used to establish the target revenue function based on the average utilization, average transmission delay, and average packet loss rate of communication channels at different quality levels. The initial allocation unit is used to determine the revenue values ​​of high-priority data, medium-priority data, and low-priority data based on the target revenue function. When all revenue values ​​are greater than the preset revenue value, the initial communication channel remains unchanged; otherwise, the initial communication channel is allocated and adjusted. The real-time adjustment unit is used to allocate and adjust the initial communication channel when it is necessary. The first priority principle is that the overall benefit value is not less than the minimum threshold. The second priority principle is to ensure the stable transmission of high-priority data, medium-priority data and low-priority data. The unit allocates communication channels for multi-level image data according to the allocation and adjustment results.

[0049] In this embodiment, the total revenue value is the sum of the revenue values ​​of priority data, medium priority data, and low priority data.

[0050] The beneficial effects of the above design scheme are as follows: By acquiring the channel parameters of the communication channel in real time, and based on the correspondence between the quality level and the channel parameters, the real-time quality level of the communication channel is determined, ensuring a high degree of matching between the channel quality level and the actual transmission state. This provides real-time and accurate level support for subsequent initial allocation and real-time adjustment. By setting differentiated channel selection priorities for high, medium, and low-level image data, it ensures that high-value core data occupies high-quality transmission resources first, while low-value redundant data does not occupy core channels, achieving hierarchical and efficient utilization of channel resources. By integrating the three core indicators of channel utilization, transmission delay, and packet loss rate, a target benefit function is constructed, transforming the rationality of allocation into a quantifiable and comparable benefit indicator. This avoids verification bias caused by qualitative evaluation and ensures the rationality and accuracy of the initial allocation scheme. Accurate judgment provides a clear quantitative standard for subsequent allocation adjustments. The benefit value of each level of data is calculated through the target benefit function and compared with the preset benefit value to accurately verify the initial allocation scheme. This avoids problems such as bandwidth waste and transmission instability caused by unreasonable allocation schemes from the source. When the benefit value is less than the preset benefit value, the allocation adjustment is initiated to ensure that the initial allocation scheme is both efficient and reasonable. The initial communication channel is allocated and adjusted according to the principle that the overall benefit value is not less than the minimum threshold as the first priority, and the second priority principle that the stable transmission of high-priority data, medium-priority data, and low-priority data is guaranteed in turn. This ensures the utilization efficiency and transmission benefit of the overall channel resources, while also prioritizing the stable transmission of high-priority core data, taking into account both the overall and individual transmission needs, and achieving stable and efficient data transmission.

[0051] Example 8: Based on Example 7, this embodiment of the invention provides a UAV dynamic detection system based on image recognition and efficient communication, wherein the target benefit function in the benefit determination unit... Specifically as follows: in, Indicates channel utilization weights, This represents the average utilization rate of all communication channels at the given quality level. Indicates the channel transmission delay weight. This represents the average transmission delay of all communication channels at the given quality level. Indicates the channel packet loss rate weight. This represents the average packet loss rate across all communication channels at the given quality level.

[0052] In this embodiment, the channel utilization weight, channel transmission delay weight, and channel packet loss rate weight can be set to, for example, 0.3, 0.3, and 0.4.

[0053] In this embodiment, when calculating the target revenue function, the average utilization rate, average transmission delay, and average packet loss rate are all pre-normalized to ensure that the calculation is reasonable.

[0054] The beneficial effects of the above design scheme are: by integrating the three core indicators of channel utilization, transmission delay and packet loss rate, a target revenue function is constructed, which transforms the rationality of allocation into a quantifiable and comparable revenue indicator, avoids the verification deviation caused by qualitative evaluation, ensures that the rationality of the initial allocation scheme can be accurately determined, and provides a clear quantitative standard for subsequent allocation adjustments.

[0055] Example 9: Based on Example 1, this embodiment of the invention provides a UAV dynamic detection system based on image recognition and efficient communication, wherein the trajectory display module includes: The receiving unit is used to receive multi-level image data transmitted by the data transmission module; The extraction unit is used to extract the target coordinates of the UAV from multi-level image data, prioritizing high-priority data, medium-priority data, and low-priority data as auxiliary data. The trajectory generation unit is used to fit a dynamic trajectory based on the target coordinates of the UAV and then visualize it.

[0056] The beneficial effects of the above design scheme are: by determining and displaying the dynamic trajectory of the UAV based on the received multi-level image data, the trajectory is calculated based on the hierarchically transmitted image data, and high-priority target data arrives and is parsed first, ensuring that the dynamic trajectory update is highly real-time, without lag or interruption, and meeting the real-time monitoring requirements of dynamic detection.

[0057] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this application and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A UAV dynamic detection system based on image recognition and efficient communication, characterized in that, include: The drone identification module is used to process and identify images collected by the drone detection terminal to obtain drone images; The image grading module is used to grade UAV images based on valid UAV information to obtain multi-level image data. The data transmission module is used to determine the encoding method and channel allocation for multi-level image data based on the real-time status of the communication channel, and to transmit the multi-level image data. The trajectory display module is used to determine and display the dynamic trajectory of the UAV based on the received multi-level image data.

2. The UAV dynamic detection system based on image recognition and efficient communication according to claim 1, characterized in that, The drone identification module includes: The repair unit is used to repair the acquired images based on the attitude data of the UAV synchronously collected by the UAV detection terminal, and to obtain the UAV repaired image based on the attitude data. The multi-scale feature extraction unit is used to extract features from UAV restored images at micro, small, medium and large scales to obtain micro-scale features, small-scale features, medium-scale features and large-scale features; The weight determination unit is used to unify the dimensions of micro-scale features, small-scale features, medium-scale features and large-scale features to obtain feature vectors at each scale. It calculates the cosine similarity between the feature vectors at each scale and the standard feature template of the UAV, determines the initial weights based on the cosine similarity, and determines the attention weights of the feature vectors at each scale based on the pre-designed spatial weight distribution map. It also determines the spatial attention weights at each scale based on the initial weights and attention weights. The weight determination unit is also used to obtain the feature vectors of 5 consecutive frames at each scale, calculate the feature similarity between the last frame and the previous 4 consecutive frames, determine the temporal attention weight of the current frame based on the similarity, and use the product of the spatial attention weight and the temporal attention weight as the comprehensive weight of the feature vector at each scale. The feature fusion unit is used to perform weighted fusion of feature vectors at all scales based on comprehensive weights to obtain initial fused features. It also analyzes the proportion of effective feature quantity in the feature vectors at each scale, retains feature vectors with an effective feature quantity proportion greater than 0.5, and suppresses feature vectors with an effective feature quantity proportion not greater than 0.5 to obtain effective feature vectors at all scales. It obtains the difference between the feature vector and the effective feature vector at each scale, and performs weighted summation of the difference based on comprehensive weights to obtain residual fused features. Finally, it adds the initial fused features and the residual fused features across all dimensions to obtain the target fused feature vector. The image determination unit is used to obtain UAV images based on target fusion feature vector integration.

3. The UAV dynamic detection system based on image recognition and efficient communication according to claim 2, characterized in that, The repair unit includes: The acquisition unit is used to acquire the angular velocity, acceleration, and heading angle of the UAV based on the attitude sensor set on the UAV detection terminal as attitude data, and to perform spatiotemporal alignment of the attitude data and the UAV image to obtain aligned attitude data; The judgment unit is used to determine the degree of blur in the UAV image based on the jitter parameters of the aligned attitude data and the mapping relationship between jitter and image blur. The image restoration unit is used to determine restoration parameters based on jitter parameters when the blur level is greater than a preset level, and to restore the acquired image based on the restoration parameters to obtain a UAV restored image.

4. The UAV dynamic detection system based on image recognition and efficient communication according to claim 2, characterized in that, The multi-scale feature extraction unit includes: The microscale feature extraction unit is used to obtain a first image from the UAV repair image with a detection distance within a first preset distance range, and extract the contour features and overall rotor features of the first image as microscale features; The small-scale feature extraction unit is used to obtain a second image from the UAV repair image with a detection distance within a second preset distance range, and extract the fuselage detail features and rotor rotation features of the second image as small-scale features; The mesoscale feature extraction unit is used to obtain a third image from the UAV repair image with a detection distance within a third preset distance range, and extract the fuselage identification features, engine features and rotor blade features of the third image as mesoscale features; The large-scale feature extraction unit is used to obtain a fourth image from the UAV repair image with a detection distance within a fourth preset distance range, and extract the fuselage structure features, rotor speed features, and landing gear features of the fourth image as large-scale features.

5. The UAV dynamic detection system based on image recognition and efficient communication according to claim 1, characterized in that, The image grading module includes: The setting unit is used to set effective information features and their corresponding priorities based on the application of historical flight images of UAVs. The level unit is used to compare the UAV image with the effective information features. When the similarity is greater than the preset similarity, the priority of the effective information features is determined as the priority of the UAV image. The data determination unit is used to obtain multi-level image data based on the priority of UAV images.

6. The UAV dynamic detection system based on image recognition and efficient communication according to claim 1, characterized in that, The data transmission module includes: The encoding setting unit is used to design a multi-layer encoding structure consisting of a base layer, a reinforcement layer, and an enhancement layer. High-priority data is encoded using the base layer, reinforcement layer, and enhancement layer; medium-priority data is encoded using the base layer and reinforcement layer; and low-priority data is encoded using the base layer, thus establishing the encoding rules. The compression ratio setting unit is used to set the coding compression ratio from low to high for the multi-layer coding structure according to the channel quality from high to low, establish the compression rules, and establish the basic coding mechanism based on the coding rules and compression rules. The channel allocation unit is used to determine the real-time quality level of the communication channel based on the real-time status of the communication channel, and to allocate communication channels for multi-level image data based on the real-time quality level. The transmission unit is used to transmit multi-level image data after encoding and compressing it according to the allocation of communication channels for multi-level image data and the basic coding mechanism.

7. The UAV dynamic detection system based on image recognition and efficient communication according to claim 6, characterized in that, The channel allocation unit includes: The level determination unit is used to obtain the channel parameters of the real-time status of the communication channel and determine the real-time quality level of the communication channel based on the correspondence between the quality level and the channel parameters. The allocation setting unit is used to set the channel selection priority of high-priority data from high to low as high-quality channel, good channel, and normal channel; the channel selection priority of medium-priority data from high to low as good channel, high-quality channel, and normal channel; and the channel selection priority of low-priority data from high to low as normal channel, good channel, and high-quality channel. The initial communication channel is allocated to multi-level image data in the above manner. The revenue determination unit is used to establish the target revenue function based on the average utilization, average transmission delay, and average packet loss rate of communication channels at different quality levels. The initial allocation unit is used to determine the revenue values ​​of high-priority data, medium-priority data, and low-priority data based on the target revenue function. When all revenue values ​​are greater than the preset revenue value, the initial communication channel remains unchanged; otherwise, the initial communication channel is allocated and adjusted. The real-time adjustment unit is used to allocate and adjust the initial communication channel when it is necessary. The first priority principle is that the overall benefit value is not less than the minimum threshold. The second priority principle is to ensure the stable transmission of high-priority data, medium-priority data and low-priority data. The unit allocates communication channels for multi-level image data according to the allocation and adjustment results.

8. The UAV dynamic detection system based on image recognition and efficient communication according to claim 7, characterized in that, The target profit function in the profit determination unit Specifically as follows: in, Indicates channel utilization weights, This represents the average utilization rate of all communication channels at the given quality level. Indicates the channel transmission delay weight. This represents the average transmission delay of all communication channels at the given quality level. Indicates the channel packet loss rate weight. This represents the average packet loss rate across all communication channels at the given quality level.

9. The UAV dynamic detection system based on image recognition and efficient communication according to claim 1, characterized in that, The trajectory display module includes: The receiving unit is used to receive multi-level image data transmitted by the data transmission module; The extraction unit is used to extract the target coordinates of the UAV from multi-level image data, prioritizing high-priority data, medium-priority data, and low-priority data as auxiliary data. The trajectory generation unit is used to fit a dynamic trajectory based on the target coordinates of the UAV and then visualize it.

Images