Unmanned aerial vehicle data acquisition and transmission system based on lossless compression technology

Abstract

The invention relates to the technical field of image communication, and discloses an unmanned aerial vehicle data acquisition and transmission system based on a lossless compression technology, which comprises a constant code rate lossy coding module, a lossless residual error generation and caching module, an entropy potential prediction module and an asynchronous compensation module, according to the system, lossless residual errors of an original frame and a reconstructed frame are stored in a cache while a lossy video stream is transmitted at a constant code rate; the method comprises the following steps of: establishing an entropy flow balancing mechanism, analyzing a motion vector field generated in a coding process to predict future information entropy, and adaptively extracting and asynchronously compensating and transmitting historical lossless residual data from a cache by utilizing a code rate surplus generated by constant code rate coding according to a prediction result. And a stable data stream matched with the channel capacity is remodeled in situ at an airborne end, so that the problem of key data loss caused by code rate impact under a weak signal is solved.

Description

Technical Field

[0001] This invention relates to a UAV data acquisition and transmission system based on lossless compression technology, belonging to the field of image communication technology. Background Technology

[0002] Currently, especially for remote data acquisition and transmission on mobile platforms such as drones, the commonly used solution is to use a high-efficiency video encoder, such as the H.264 or H.265 standard, to compress the raw high-definition video captured by the airborne camera, and then transmit it to the ground receiving end via a wireless channel. The core technical principle is that the data bitrate output by the video encoder is not a fixed value, but has a direct physical relationship with the complexity of the image content itself, i.e., information entropy. When the drone lens sweeps over a simple sky or water surface, the image information entropy is low, and the bitrate output by the encoder also decreases accordingly. However, when the lens turns to a detailed forest or building, the information entropy increases sharply, and the encoder will inevitably output a high bitrate data stream instantaneously to ensure image quality.

[0003] However, when this technology is applied to long-endurance, long-distance industrial inspection tasks, especially when conducting power line or pipeline surveys in mountainous or remote areas, a fundamental engineering dilemma emerges. The wireless communication links that UAVs rely on typically have weak signals and a consistently low bandwidth. This leads to an irreconcilable physical mismatch between the dynamically fluctuating entropy flow generated by the encoder and the inherently relatively constant pipeline capacity of the communication channel. In this best-effort transmission mode, once the UAV enters a critical survey area with high information entropy, the instantaneous bit rate peak generated by the encoder can easily exceed the channel capacity limit. The direct consequence is the rapid filling and overflow of the onboard transmission buffer, causing a large amount of data, which usually contains the most critical details, to be permanently discarded at the transmitting end. This results in structural damage and widespread image errors in the video stream received at the ground end, making it impossible to guarantee the reliability and data integrity of the inspection operation.

[0004] To alleviate this problem, those skilled in the art have explored various approaches. One approach attempts to use more complex bitrate control algorithms to predict and reduce encoding quality in advance to avoid bitrate exceeding limits. However, this method relies on unreliable predictions and sacrifices image clarity at critical moments. Another approach involves configuring large-capacity physical storage devices on the airborne end for storage before transmission. However, this increases the weight, power consumption, and system complexity of the drone, violating the core design requirements of lightweight and long-endurance drone platforms, and sacrificing real-time operation. Analysis reveals that existing technical approaches have the following shortcomings: 1. They treat the encoder as an independent entropy source and the channel as an independent pipeline, lacking an inherent collaborative mechanism that enables adaptive matching between information flow and channel capacity; 2. Under the inevitable impact of the physical mismatch between high-entropy data and channel capacity, existing systems lack a resilient structure capable of preserving data integrity and preventing catastrophic information loss. Therefore, the technical problem to be solved by this invention is how to devise a way for UAV data acquisition and transmission that can reshape the uneven entropy stream generated by video encoding into a smooth and stable data stream on-site at the airborne end, so as to match it with the fragile constant bandwidth channel at the physical level, thereby avoiding the loss of critical data due to buffer overflow. Summary of the Invention

[0005] This invention provides a UAV data acquisition and transmission system based on lossless compression technology. Its main purpose is to solve the problem of massive loss of key information during data transmission in weak network environments due to the physical mismatch between the information entropy fluctuation of image content and the constant bandwidth of the channel.

[0006] To achieve the above objectives, the present invention provides a UAV data acquisition and transmission system based on lossless compression technology, the system comprising: A constant bitrate lossy encoding module specifically encodes the original video frame sequence captured by the UAV into a lossy video stream based on a constant target bitrate, and generates the corresponding reconstructed video frame sequence. A lossless residual generation and caching module specifically calculates the pixel difference between the original video frame and the reconstructed video frame frame by frame to generate lossless residual data associated with the frame identifier, and stores the lossless residual data in an onboard buffer in frame order. An entropy potential prediction module specifically extracts the motion vector field generated by the constant bit rate lossy coding module in real time and calculates the statistical parameters of the motion vector field to generate a prediction signal characterizing the future information entropy trend. An asynchronous compensation module specifically comprises: receiving a prediction signal; calculating the difference between the constant target code rate and the actual coding code rate of the constant code rate lossy coding module to determine the code rate surplus; and adjusting the rate at which lossless residual data is retrieved from the head of the airborne buffer based on the prediction signal, and using the channel bandwidth corresponding to the code rate surplus to transmit the retrieved lossless residual data.

[0007] Preferably, the asynchronous compensation module further encapsulates the extracted lossless residual data into a compensation data packet carrying the frame identifier associated with it, and embeds the compensation data packet into the lossy video stream being transmitted, so as to transmit synchronously within a single data channel.

[0008] Preferably, the entropy potential prediction module calculates the statistical parameters of the motion vector field, specifically by calculating the variance of the angles of all motion vectors in the motion vector field. The calculation rule is as follows ,in, The total number of motion vectors in the current frame. For the first The angle of each motion vector Let be the average of the angles of all motion vectors; and, when the variance When the value is lower than the first threshold, the prediction signal instructs the asynchronous compensation module to reduce the rate of retrieving lossless residual data.

[0009] Preferably, the system further includes a ground-end fusion and reconstruction module, which specifically performs the following functions: receiving and separating the lossy video stream and compensation data packets from the transport stream; decoding the lossy video stream; and, based on the frame identifier carried by the compensation data packet, pixel-level superimposing the lossless residual data with the decoded lossy video frame corresponding to the frame identifier, so as to restore the lossy video frame to a lossless video frame.

[0010] Preferably, the system further includes an adaptive region of interest (ROI) generation module, which specifically performs the following steps: calculating the energy values ​​of all pixel macroblocks in the lossless residual data within a single video frame to generate a residual energy map; calculating the global average energy value of the residual energy map; and defining the coordinates of connected regions on the residual energy map whose energy values ​​are higher than the global average energy value multiplied by a coefficient as ROI metadata and transmitting them.

[0011] Preferably, the lossless residual generation and caching module further comprises: decomposing the lossless residual data of each frame into a high-priority structural residual part containing low-frequency coefficients and a low-priority texture residual part containing high-frequency coefficients through integer discrete cosine transform and two-level quantization processing; and the system further comprises a cache manager, which specifically discards the corresponding low-priority texture residual part only from the lossless residual data stored earliest in the airborne cache when the occupancy rate of the airborne cache exceeds a cache pressure threshold.

[0012] Preferably, the system further includes an image quality assessment and priority calibration module, which specifically monitors the sequence of quantization parameter QP values ​​within the constant bitrate lossy coding module; when it is detected that the decrease in the quantization parameter QP value exceeds a QP change threshold within a single time window, the corresponding original video frame is judged as a low-quality frame; and when the lossless residual data generated from the low-quality frame is stored in the onboard buffer, a low recovery priority is assigned so that the asynchronous compensation module prioritizes the transmission of lossless residual data with high recovery priority.

[0013] Preferably, the system further includes a pre-filtering module, which is located before the original video frame sequence enters the constant bit rate lossy encoding module and the lossless residual generation and caching module. Specifically, the pre-filtering module performs a fast Fourier transform on the original video data based on spectral data recording the vibration frequency characteristics of the UAV to convert it to the frequency domain. In the frequency domain data, a digital notch filter is used to suppress the frequency band energy corresponding to the vibration frequency characteristics, and then an inverse fast Fourier transform is used to restore the purified spatial domain image data.

[0014] Preferably, the system also includes a navigation complexity map generation module, which specifically: obtains the real-time occupancy rate of the airborne buffer at a fixed time frequency; reads the geographical coordinates of the airborne global positioning system module at the same time; and associates the real-time occupancy rate with the geographical coordinates to generate map data representing the information complexity of different geographical locations on the flight path.

[0015] Preferably, the onboard cache area in the lossless residual generation and caching module is a circular cache based on random access memory, which follows the first-in-first-out data read and write rule.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention establishes a dynamic balancing mechanism for image data streams. It generates a stable reference video stream through a constant bitrate lossy encoding module and utilizes the channel space corresponding to the padding data output by this module to maintain a constant bitrate when processing low-complexity images to transmit the high-information lossless residual data that was previously backed up in the buffer due to processing high-complexity images. This mechanism transforms the naturally non-uniform temporal information entropy distribution of image content into a uniform and stable data stream output on the transmission channel. Without relying on external channel quality prediction, it enables UAVs to simultaneously provide uninterrupted real-time situational awareness images and post-event bit-accurate lossless video recordings to the ground under signal-constrained communication conditions.

[0017] 2. Before performing asynchronous residual compensation, the statistical characteristics of the motion vector field generated during lossy encoding are analyzed. When the vector field tends to be consistent, the system judges that the information entropy of the future image will decrease, and then actively adjusts the behavior of the asynchronous compensation module to reserve buffer space for the upcoming high information entropy impact. This method transforms a passive response entropy flow balancing process into a closed-loop control system with predictive and proactive adjustment capabilities, which structurally enhances the system's data preservation capabilities and its ability to absorb sudden high entropy events when dealing with tasks with clear transitions from simple to complex scenarios.

[0018] 3. By correlating the internal state parameters of the encoding process with the acquisition quality of the external image, and monitoring the quantization parameters dynamically adjusted by the constant bitrate encoder to maintain the bitrate, image quality degradation events caused by motion blur or defocus are identified. Based on this, a lower recovery priority is assigned to the lossless residual data generated for the corresponding frame. This allows the asynchronous compensation module to prioritize residuals associated with high-quality acquired images when using bitrate surplus for data transmission, allocating limited channel resources to video frames with higher information value. This avoids the transmission and storage of invalid information and allows for the reuse of byproduct information during data transmission. By analyzing the spatial energy distribution of lossless residual data within video frames in real time, the system identifies image regions with localized high-energy residuals caused by prediction failures and sends these regions as metadata of the region of interest to the ground. Simultaneously, the system also maps the real-time occupancy rate of the entropy cache pool with the UAV's geographical location information to generate map data representing the information complexity along the flight path. These two methods transform a simple data preservation system into an information assistance tool that combines front-end real-time perception with back-end intelligent operational planning capabilities, providing users with data insights and decision-making dimensions without increasing sensors or major computing load. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall functional architecture and data flow of the system of the present invention; Figure 2 This is a logic diagram showing the core working state transition of the asynchronous compensation module of the present invention; Figure 3 This is a time series diagram of the prospective compensation rate adjustment based on entropy potential prediction in this invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention provides a UAV data acquisition and transmission system based on lossless compression technology. Its overall architecture consists of an airborne processing unit deployed on the UAV and a ground-based receiving and reconstruction unit. The airborne processing unit mainly integrates a constant bitrate lossy encoding module, a lossless residual generation and caching module, an entropy potential prediction module, and an asynchronous compensation module. The ground-based receiving and reconstruction unit integrates a ground-based fusion and reconstruction module. During system operation, the airborne processing unit first generates and transmits a lossy video stream with a stable bitrate in real time through the constant bitrate lossy encoding module to ensure basic remote situational awareness. In parallel, the lossless residual generation and caching module converts the original data into a lossy video stream. The difference between the original video and the lossless reconstructed video, i.e., the lossless residual, is stored in an onboard buffer. Subsequently, the entropy potential prediction module predicts the information entropy trend of future images by analyzing the motion vector field, a byproduct of the lossy coding process. Based on this prediction, the asynchronous compensation module dynamically utilizes the channel bandwidth corresponding to the bit rate surplus generated by lossy coding in low-entropy scenarios to asynchronously compensate and transmit the lossless residual data of historical high-entropy scenarios accumulated in the buffer to the ground end. Finally, the ground-end fusion and reconstruction module receives all the data and completes the restoration of the lossless video, thereby achieving uninterrupted real-time observation and post-event bit-accurate full-path data recording under weak signal image communication links.

[0022] In a specific application scenario, the system is deployed on a multi-rotor drone performing long-distance power line inspection missions. This drone needs to traverse mountainous areas where signal strength fluctuates between -95dBm and -105dBm, and its available uplink bandwidth is stably assessed at 2Mbps. To cope with this type of weak network environment common in H04N image communication, the constant bit rate lossy encoding module is configured as an H.265 hardware encoder with a strictly constant target bit rate of 1.8Mbps upon system startup. This value was determined by pre-setting a 10% safety margin after considering the stability of the channel bandwidth and signaling overhead. This module encodes the 1080p raw video frame sequence captured by the airborne camera into a lossy video stream and transmits it in real time via a wireless link. This establishes a fixed reference data channel, ensuring that ground operators can always receive smooth base video regardless of subsequent scene changes. At the same time, the lossless residual generation and caching modules work in parallel. The module calculates the pixel difference between the raw video frame and the reconstructed video frame output by the constant bitrate lossy encoding module frame by frame, generates lossless residual data associated with the timestamp or frame number of that frame, and stores this data in a 256MB circular buffer built on the airborne processor RAM according to the first-in-first-out rule. This buffer is the airborne buffer, which is used for temporary data storage when the information generation rate does not match the channel transmission rate.

[0023] As the drone approaches a forest from an open area, to prevent the onboard buffer from being rapidly filled without preparation due to a sudden increase in information entropy, the entropy potential prediction module is activated. This module utilizes the principle that the macroscopic statistical characteristics of motion vector fields in video coding technology can reflect the complexity and motion trend of image content. It extracts the motion vector field data generated by the constant bitrate lossy coding module during inter-frame prediction in real time and calculates the variance of the angles of all motion vectors in this vector field. Its calculation procedure is set as a formula. ,in, The total number of motion vectors in the current frame. For the first The angle of a motion vector, and This is the average angle of all motion vectors; when the drone flies towards a distant, uniformly textured sky, the motion vectors in different areas of the image tend to be consistent. For example, in a sample containing 100 motion vectors, the angles are... arrive The calculated values ​​fluctuate slightly between these values. The value is 2.5. However, when the drone approaches the forest, the parallax effect causes the pixel blocks in the foreground and background to move in completely different directions, and the motion vector angles are distributed in... arrive The interval, calculated The value will then rise above 800. Based on this, the system has a built-in calibrable first threshold. This threshold is determined by offline analysis of a large number of inspection video samples containing scene changes, and statistical analysis of their values. The distribution pattern of values ​​before and after scene switching is analyzed, and a value that can distinguish between stable flight and approaching complex areas with a 95% confidence level is selected, such as 50. When the entropy potential prediction module detects... When the value of the signal remains below the first threshold, it generates a prediction signal representing future low entropy and sends it to the asynchronous compensation module; conversely, it generates a prediction signal representing future high entropy. The asynchronous compensation module, as the system's execution mechanism, operates under the combined influence of the generation of the rate surplus and the entropy potential prediction signal. When a drone flies over low-entropy scenarios such as skies or farmland, the constant rate lossy encoding module, to maintain a constant output of 1.8 Mbps, will pad the remaining 1.0 Mbps bandwidth by filling in data after completing image information encoding (at which point only 0.8 Mbps may actually be consumed). The asynchronous compensation module detects this portion of channel capacity that will be used for padding and defines it as the rate surplus that can be used to compensate for transmission. At this point, it adjusts its data compensation strategy based on the received prediction signal. If the prediction signal represents future low entropy... Entropy indicates that subsequent channel resources are relatively abundant. The module will retrieve the earliest stored lossless residual data of the previously backed-up forest scene from the head of the airborne buffer at the rate corresponding to the current full bitrate surplus (1.0Mbps). Conversely, if the predicted signal is high entropy in the future, the module will actively reduce the retrieval rate, for example, only using 50% of the bitrate surplus (i.e., 0.5Mbps) for compensation. The purpose is to slow down the emptying speed of the buffer, thereby reserving more buffer space for the upcoming high entropy data impact. The retrieved lossless residual data will be encapsulated into compensation data packets carrying their original frame identifiers and embedded into the lossy video stream being transmitted. It will be sent synchronously through a single data channel. This working method of adjusting the compensation rate based on the predicted signal enables the system to have the ability to adjust in advance to cope with scene changes.

[0024] It should be noted that the asynchronous compensation module embeds the compensation data packet into the lossy video stream using the Supplemental Enhancement Information (SEI) message mechanism provided by modern video coding standards such as H.265. Specifically, the asynchronous compensation module first encapsulates the lossless residual data carrying the frame identifier into a custom data structure and adds a globally unique 128-bit identification code to its header to identify the system of this invention. Subsequently, the entire data structure is written as the payload of a Supplemental Enhancement Information (SEI) message and inserted after the inter-frame prediction image in the video stream output by the constant bitrate lossy coding module. Correspondingly, when the ground-side fusion and reconstruction module parses the bit stream, it will launch an SEI message scanner in parallel. This scanner parses all SEI messages in the bit stream one by one and accurately identifies and extracts the compensation data packet dedicated to this system by comparing the 128-bit identification code in the header, and then sends it to the subsequent fusion and reconstruction process. This method utilizes the existing framework of video coding standards without compromising the system's performance. Under the premise of the main syntax structure of the video stream, reliable transmission and separation of heterogeneous data within a single channel are achieved. When the mixed data stream arrives at the ground end, the ground end fusion and reconstruction module is responsible for parsing and recovery. This module first separates the lossy video stream from the compensation data packet based on the identifier of the data packet. The lossy video stream is sent to the standard decoder and displayed on the operator's monitoring screen after real-time decoding to ensure the immediacy of the task. The separated compensation data packet is stored in a temporary buffer. After the decoder completes the decoding of a lossy image frame, it checks whether the identifier of the frame matches the frame identifier of a compensation data packet in the temporary buffer. If the match is successful, the lossless residual data in the compensation data packet is arithmetically superimposed pixel by pixel with the decoded lossy video frame, thereby restoring the image frame to the original lossless video frame at the bit level and storing it on the hard disk for detailed analysis after the task is completed.

[0025] To further enhance the system's applicability and resource utilization efficiency, this solution can also integrate a series of functional modules. Among them, the adaptive region of interest generation module uses lossless residual data to generate operation prompts. Its working principle is that the energy of the lossless residual directly reflects the difficulty of predicting the original image. A broken insulator or a small landslide, compared to its surrounding background, is a region that is difficult to accurately describe by inter-frame or intra-frame prediction models, resulting in high-energy residuals at its corresponding locations. In view of this, the module calculates the energy value (i.e., the sum of squares of the residuals) of the lossless residual data of each frame in real time on the airborne end, using 16x16 pixel macroblocks as units, and generates a residual energy map of the same size as the original image. Subsequently, the module calculates the global average energy value of the map, for example, 500, and multiplies all energy values ​​higher than this average by a preset coefficient (for example, 3).The coordinates of a connected region (i.e., 1500) are defined as the region of interest metadata. This metadata is sent to the ground terminal, where the corresponding region is highlighted or highlighted on the operator's interface to assist the operator's visual focus. Furthermore, to cope with continuously high-entropy scenarios such as drones flying over vast forests, the lossless residual generation and caching module can integrate a residual data layering and selective unloading mechanism. After generating the lossless residual, this mechanism uses integer discrete cosine transform and two-level quantization to decompose the residual of each frame into low-frequency coefficients containing image contour information (high-priority structural residuals) and high-frequency coefficients containing fine texture information (low-priority texture residuals), which are then stored together in the onboard cache. The system's cache manager continuously monitors the occupancy rate of the onboard cache and sets a cache pressure threshold, such as 90%. Once the occupancy rate exceeds this threshold, when the cache manager needs to free up space for new data, it no longer discards the entire oldest residual data frame, but selectively discards only the associated low-priority texture residuals from that frame. This prioritizes data preservation when cache space is insufficient. The system includes the core structural information of all video frames. It can also include an image quality assessment and priority calibration module to optimize channel bandwidth allocation when transmitting low-value image data caused by physical jitter or defocusing. This module utilizes the quantization parameter QP value, dynamically adjusted internally by the constant bitrate lossy coding module to maintain a constant bitrate. The QP value reflects the difficulty of the encoder compressing an image. An image with motion blur has high internal redundancy, and the encoder only needs a lower QP value to complete the compression, for example, QP=15. In contrast, a clear image... High-quality images require a higher QP value, such as QP=35. This module monitors the sequence of QP values. When it detects that the QP value drops by more than a preset QP change threshold within a single time window (e.g., 5 frames), such as from 35 to 15 with a change exceeding 50%, the corresponding original video frame is classified as a low-quality frame. The lossless residual data generated for this frame is assigned a low recovery priority when stored in the onboard buffer. Correspondingly, when the asynchronous compensation module retrieves data, it prioritizes transmitting the lossless residual data with a high recovery priority.

[0026] To optimize the quality of the data source entering the system, a pre-filtering module can also be configured. This module is located before the raw video frame sequence enters the constant bitrate lossy encoding module and the lossless residual generation and caching module. Its function is to suppress high-frequency vibration noise caused by the UAV motor and propellers. This noise manifests as pixel-level distortion in the image and is interpreted as complex texture by the encoder, thus generating a large amount of useless residual data to fill the cache space. This module uses spectral data pre-calibrated through ground testing, recording the vibration frequency characteristics of this UAV model (e.g., the main vibration frequencies are concentrated at 150Hz and 180Hz), to perform a Fast Fourier Transform on each frame of raw video data to convert it to the frequency domain. In this frequency domain data, one or more digital notch filters are used to suppress the frequency band energy corresponding to the vibration frequency characteristics. Then, an Inverse Fast Fourier Transform is used to recover the purified spatial domain image data. This improves the quality of the lossy reference stream. This reduces the impact of invalid residual data on cache space. Finally, to provide data analysis support for UAV inspection operations, the system can also integrate a navigation complexity map generation module. This module uses the real-time occupancy rate of the airborne cache as a measure of the information complexity on the flight path. It synchronously acquires the real-time occupancy percentage of the airborne cache and the geographic coordinates provided by the airborne GPS module at a fixed time frequency, such as once per second, and associates these two data to generate a data tuple, such as Timestamp: 1662000000, GPS: (36.66, 117.02), Occupancy: 85%. These tuple data are transmitted back to the ground or exported uniformly after the mission. On the ground-based geographic information system software, a navigation complexity map representing the information complexity of different geographic locations on the flight path can be generated, providing data reference for subsequent mission review and route planning.

[0027] Example 1: In an emergency structural survey of a key bridge in a mountainous area after an earthquake, a drone equipped with the system of this invention was deployed. Its mission objective was to collect high-resolution visual data of the core load-bearing cable connection of a suspension bridge located in a valley with weak signal and surrounded by dense vegetation. Testing showed that the available uplink image communication link bandwidth between the drone and the ground station was a constant 2Mbps. In this scenario, conventional image transmission methods face a technical bottleneck: if the encoding bitrate is set below 2Mbps to ensure continuous transmission, the image quality will deteriorate due to over-compression when capturing close-ups of the cable connection, resulting in the loss of structural crack details. If a higher variable bitrate encoding is used to preserve details... When the drone's camera scans the bridge and surrounding vegetation, the instantaneous bit rate peak generated by the encoder will exceed the channel limit of 2Mbps, causing the onboard transmission buffer to overflow. This results in a significant loss and structural damage of critical bridge structure image data. After the drone takes off, the system activates a constant bit rate lossy encoding module with a target bit rate of 1.8Mbps to continuously transmit lossy video streams to the ground, thus maintaining the pilot's basic perception of the drone's position and attitude. As the drone approaches the vegetation-covered bridge area from a relatively open riverbed along the predetermined route, the entropy potential prediction module analyzes the motion vector field generated by the constant bit rate lossy encoding module and detects a decrease in the consistency of motion vectors of objects in the image, with its statistical variance... The value climbed from 10.5 when flying over the river to 65.8 within 3 seconds, exceeding the preset first threshold of 50. Based on this, the system generated a prediction signal for future high entropy. After receiving this signal, the asynchronous compensation module, even though the drone was still in an open area and there was still a code rate surplus in the channel, immediately reduced the rate at which it extracted and sent lossless residual data from the airborne buffer from 100% of the available surplus to 20%. Through this pre-adjustment, the available space ratio of the airborne buffer was maintained at a high level.

[0028] Subsequently, the drone arrived at the bridge location and hovered to take pictures of the cable connection. At this moment, the camera lens captured details such as metal corrosion and tiny cracks in the concrete, causing a sudden increase in image information entropy. The peak rate of the lossless residual data stream generated by the lossless residual generation and caching modules reached 11.5 Mbps. This data, containing structural damage information, was written to the onboard buffer at high speed. Due to the coordinated operation of the entropy potential prediction module and the asynchronous compensation module in the previous stage, the buffer reserved absorption space. This high-entropy data surge caused by the key survey target was completely received by the onboard buffer without overflow. During this period, the 1.8 Mbps lossy video stream received by the ground operator remained smooth, and the drone's flight near the bridge structure was stably controlled. After completing the key data acquisition, the drone returned and flew over the open river again. The entropy potential prediction module detected the variance of the motion vector field. The asynchronous compensation module, after dropping back to 12.0, resumed its maximum compensation rate. Utilizing the resulting channel rate surplus, it asynchronously compensated and transmitted the lossless residual data from the bridge cable connections previously stored in the airborne buffer to the ground station in batches. After the mission, the ground-based fusion and reconstruction module fused all received compensation data packets with the corresponding lossy video frames to generate a complete lossless video record of the entire survey mission. Data verification confirmed that the record contained all high-resolution images taken by the UAV at the bridge cable connections, and its data was bit-level consistent with the original data acquired by the airborne station, providing a data foundation for subsequent engineering structure analysis. In this workflow, the system reshaped the non-uniform information entropy stream generated by the encoder in the time dimension into a uniform data stream on the transmission channel through airborne buffering and scheduling, thus achieving both real-time situational awareness and post-event data integrity under a fixed bandwidth channel.

[0029] Example 2: To quantitatively verify the adaptability and data preservation capability of the technical solution of this invention to a fixed bandwidth channel under the impact of video streams with fluctuating information entropy, a hardware-in-the-loop simulation-based test platform was built. This test platform consists of a computing unit responsible for simulating UAV onboard data processing and a computing unit responsible for simulating the ground receiver. The two units establish a communication link through a network simulator configured to simulate a wireless communication channel with a bandwidth limit of 2Mbps and a transmit buffer of 512KB. These channel parameters are used to reproduce the weak signal narrowband transmission environment faced by UAVs performing missions in remote areas. The input data used in the experiment comes from a standard test video sequence for video encoder performance evaluation publicly released on the Xiph.org website. A 60-second video segment with a resolution of 1080p was selected from this sequence. The content of this segment... The experiment simulated a simple-complex-simple scene transition process: the first 20 seconds depicted a low-information-entropy scene of skyward movement, the middle 20 seconds a high-information-entropy scene of swaying leaves in the woods, and the last 20 seconds a return to a low-information-entropy scene of skyward movement. A control group and an experimental group were set up for performance comparison. The control group used a common technique in H04N image communication, namely an H.255 video encoder operating in variable bitrate mode with an average target bitrate of 1.8 Mbps. The encoded data stream was directly sent to the 512KB transmission buffer of the network simulator. The experimental group fully deployed the technical solution of this invention, with its constant bitrate lossy encoding module having a target bitrate of 1.8 Mbps and an onboard buffer capacity of 256 MB. During the experiment, the overflow packet loss rate of the instantaneous bitrate transmission buffer generated on the airborne end and the integrity of the finally recovered video data on the ground end were recorded in real time for both groups.

[0030] At the 21-second mark of the experiment, the video content switched from a low-information-entropy sky to a high-information-entropy forest. To maintain image quality, the onboard encoders in both groups increased the instantaneous bitrate to 9.8 Mbps. In the control group, this 9.8 Mbps data stream directly impacted the transmit buffer, causing it to fill up and overflow within 200 ms. During the entire high-information-entropy period, the average overflow packet loss rate of the transmit buffer was as high as 79.6%. The video stream received by the ground station during this period showed widespread decoding errors, making the image unrecognizable. In contrast, in the experimental group, the lossless residual data corresponding to the 9.8 Mbps instantaneous bitrate was completely stored in the onboard buffer, causing its occupancy rate to steadily increase from 10.1%, while the output of its constant bitrate lossy encoding module remained at 1.8 Mbps. Therefore, the transmit buffer... The occupancy rate did not exceed 90%, the overflow packet loss rate was 0, and the lossy video stream received by the ground end remained smooth throughout. At the 41st second of the experiment, after the video content returned to a low-entropy scene, the asynchronous compensation module of the experimental group used the bitrate surplus to compensate and transmit all the residual data stored in the airborne buffer to the ground end within 15 seconds. The buffer occupancy rate then dropped, and finally, the ground end fusion and reconstruction module recovered the bit-accurate lossless video. The experimental data shows that when faced with the impact of increased video source information entropy, the control group experienced data loss due to the lack of a bitstream peak management mechanism, while the experimental group, through airborne buffering and asynchronous compensation, smoothed the instantaneous data peaks generated in the high-entropy scene in the time domain and used the channel capacity of the low-entropy scene to complete the transmission, thus achieving complete preservation of video data without exceeding the physical capacity limit of the channel.

[0031] Example 3: This example combines Figures 1 to 3 This describes a UAV data acquisition and transmission system based on lossless compression technology, such as... Figure 1As shown, its workflow begins with raw video frames captured by the airborne camera. These raw video frames are sent in parallel to two processing paths. In the first path, the constant bitrate encoding module encodes the raw video frames into a lossy video stream and transmits it to the fusion and reconstruction module on the ground, while simultaneously generating reconstructed frames. In the second path, the lossless residual generation module generates lossless residual data based on the raw video frames and reconstructed frames and stores it in the D1 airborne buffer. At the same time, the motion vector field generated by the constant bitrate encoding module is sent to the prediction information entropy module. After analysis, this module generates a prediction signal and sends the prediction signal and the bitrate surplus information generated by the constant bitrate encoding module to the asynchronous compensation transmission module. The asynchronous compensation transmission module retrieves the residual data to be transmitted from the D1 airborne buffer based on the received signal, encapsulates it into a compensation data packet, and sends it to the fusion and reconstruction module on the ground. After receiving the lossy video stream and the compensation data packet, the fusion and reconstruction module outputs real-time images for ground operators to monitor, and simultaneously fuses and reconstructs the two into a lossless video record and archives it.

[0032] like Figure 2 As shown, this module has three core working states. Its initial state is standby, in which there is no bitrate surplus or the buffer is empty. When a bitrate surplus occurs and a future low-entropy signal is received, the module will transition from standby to full-speed compensation, using all bitrate surplus to transmit the residual. If a future high-entropy signal is received in full-speed compensation, the module will transition to deceleration compensation, where some surplus will be reserved as a buffer for high-entropy impact. Conversely, if a future low-entropy signal is received in deceleration compensation, the module will transition back to full-speed compensation. In addition, in standby, if a bitrate surplus occurs but a future high-entropy signal is received, the module will directly enter deceleration compensation. In full-speed compensation or deceleration compensation, once the bitrate surplus disappears or the buffer is empty, the module will return to standby until the task ends, and the entire workflow terminates.

[0033] like Figure 3 As shown, the interaction process begins with the constant rate encoding module providing motion vector field data to the entropy potential prediction module. The entropy potential prediction module then calculates the vector angle variance and makes a judgment based on a preset threshold. When the variance is lower than the threshold, a future low-entropy prediction signal is generated; when the variance is higher than the threshold, a future high-entropy prediction signal is generated. This prediction signal is sent to the asynchronous compensation module. At the same time, the constant rate encoding module also sends the rate surplus information to the asynchronous compensation module. The asynchronous compensation module determines its compensation rate based on the received prediction signal. If it is a future low-entropy signal, the compensation rate is increased to 100%; if it is a future high-entropy signal, the compensation rate is decreased to 50%. Based on the adjusted rate, the residual is retrieved from the onboard buffer, encapsulated into a compensation data packet, and sent via the transmission channel by embedding an SEI message.

[0034] Example 4: Before applying the technical solution of this invention to a new model of UAV platform, in order to adapt the working parameters of its internal functional modules to the characteristics of the platform's image sensor and flight dynamics, an offline calibration procedure needs to be performed. This procedure determines the working values ​​of the key judgment thresholds in the system through a reproducible engineering step, so that the data preservation efficiency and operational stability of the system meet the design requirements. The initial conditions for calibration are a UAV hardware platform equipped with the system to be calibrated, and a set of pre-recorded 1080p high-definition video datasets as standard input signals. The dataset consists of four video segments: the first segment is a low-information-entropy baseline video of the UAV flying smoothly over an open water surface; the second segment is a high-information-entropy baseline video of the UAV flying low over a forest; the third segment is a scene switching video of the UAV flying towards a complex building complex after takeoff; and the fourth segment is a sample video of image quality degradation caused by several rapid nose rotations during normal flight.

[0035] The calibration process first determines the first threshold used for the entropy potential prediction module. During operation, low-information-entropy and high-information-entropy reference videos are input into the system respectively, and the entropy potential prediction module is run in analysis mode to record the statistical variance of the motion vector field corresponding to each frame of the image. The values, after data collection was completed, were analyzed for both groups. Numerical statistical analysis was performed, among which low information entropy videos... The numerical set has a mean of 11.2 and a standard deviation of 3.5, while the high information entropy video... The set of values ​​has an average of 680.4. The first threshold is calculated by adding three standard deviations to the average of the low-information-entropy set of values. The result is... The threshold was rounded down and set to 25. Then, the QP change threshold for the image quality assessment and priority calibration module was determined. During operation, a sample video of degraded image quality was input into the system, and the QP value sequence of the constant bitrate lossy encoding module was recorded frame by frame during the encoding process. By comparing the video footage with the QP value data, it was found that when the drone was flying smoothly, the QP value fluctuated between 28 and 34. However, when performing a rapid nose-rotation maneuver, the QP value dropped from 32 to 16 within 3 frames in the initial frame where motion blur appeared. Based on this, the QP change threshold was determined. The judgment logic for the value is set as follows: within a time window of 5 consecutive frames, if the QP value of the current frame decreases by more than 50% relative to the highest QP value in the window, a low-quality frame judgment is triggered. Therefore, the QP change threshold is set to 50%. By executing this calibration procedure, the judgment criteria of key modules in the system are all based on measured data. This procedure enables a general technical solution to be adapted to specific hardware platforms and application scenarios, reducing the probability of affecting system performance due to improper parameter settings, and providing an engineering basis for the stable operation of the system during actual deployment.

[0036] Example 5: Before integrating the system of the present invention into a specific model of UAV and deploying it for the first time, in order for the pre-filter module to suppress the vibration noise generated by the UAV's airframe structure and power system during flight, a field calibration procedure for vibration frequency characteristics needs to be performed. This procedure fixes the UAV on a test bench and simulates the motor speeds of three working states—hovering slow cruise and fast maneuvering—in sequence under no load. Using an onboard triaxial accelerometer with a sampling rate of 200Hz, the vibration data of the airframe under each working state is recorded synchronously. Subsequently, by performing a fast Fourier transform on the collected acceleration data samples of each axis, the narrowband frequency peaks with concentrated energy on the spectrum are identified. These frequency values, 155Hz and 182Hz, and their corresponding energy amplitudes are recorded as the vibration frequency characteristic spectrum data of the UAV model and fixed in the parameter table of the pre-filter module as the basis for its targeted frequency filtering.

[0037] When the system performs a long-term survey of vast post-disaster ruins, it faces continuous high information entropy input. This causes the onboard buffer occupancy rate to remain above 90% of the buffer pressure threshold even after triggering the residual data stratification and selective offloading mechanism and discarding all low-priority texture residuals. Under these conditions, to prevent high-priority structural residual data containing structural information from being discarded due to buffer overflow, the buffer manager sends a control command to the constant bitrate lossy encoding module, increasing its internal quantization parameter QP value by 2 steps. This adjustment slightly reduces the image quality of the real-time lossy video stream, but its effect is to reduce the difference between the lossy encoded reconstructed frame and the original frame, thereby reducing the amount of newly generated lossless residual data. This mechanism reduces the pressure on the onboard buffer by adjusting the fidelity of the real-time image, thus mitigating the risk of historical data loss due to buffer overflow under continuous entropy saturation.

[0038] Example 6: To ensure that the adaptive region of interest generation module meets engineering requirements for fault indication accuracy and false alarm rate in power line inspection tasks, its core judgment parameter, namely the segmentation coefficient of the residual energy map, needs to be optimized and calibrated offline. The objective function of this calibration procedure is to find a segmentation coefficient value that maximizes the fault identification rate while controlling the false alarm rate below a preset level. The procedure uses a video dataset containing positive and negative samples. The positive samples are several video segments containing known insulator damage, while the negative samples are videos with a total duration of more than 10 hours recording various normal poles. For the inspection videos of towers and terrain, the calibration process started with a segmentation coefficient of 2.0. The entire video dataset was processed, and the number of accurately identified fault points and the number of false alarms in normal videos were counted. Then, the segmentation coefficient was increased in increments of 0.5 and the statistical process was repeated until the coefficient reached 5.0. Finally, the fault identification rate and the false alarm rate per unit hour under different coefficient values ​​were analyzed. It was determined that when the segmentation coefficient was 3.5, the fault identification rate reached 96% and the false alarm rate per unit hour was less than 1. This coefficient value was determined as the optimal working point for this inspection task scenario and was fixed in the system configuration.

[0039] To address the potential momentary loss or accuracy degradation of GPS signals during mountain patrols, the system integrates a set of data validity verification and marking rules during pre-deployment verification of the navigation complexity map generation module. These rules stipulate that before recording data tuples, the module queries the Horizontal Precision Factor (HDOP) value of the airborne GPS module. Only when the HDOP value is below the validity threshold of 2.0 is a complete data tuple containing timestamps, geographic coordinates, and buffer occupancy rate recorded. If the HDOP value is above this threshold, the module pauses recording geographic coordinates and instead records a data tuple containing only timestamps, buffer occupancy rate, and an invalid geographic location identifier. This rule prevents inaccurate geographic location information from being written into the map data. When rendering the navigation complexity heatmap, the ground application software uses this invalid identifier to mark the corresponding time period with dashed lines, thus presenting areas with poor signal on the map while simultaneously retaining continuous records of information complexity within those areas.

[0040] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An unmanned aerial vehicle data acquisition and transmission system based on lossless compression technology, characterized in that, The system comprises: a constant-rate lossy encoding module, which specifically encodes the original video frame sequence collected by the UAV into a lossy video stream according to a constant target code rate, and generates a corresponding reconstructed video frame sequence; a lossless residual generation and caching module, which specifically calculates the pixel difference between the original video frame and the reconstructed video frame to generate lossless residual data associated with the frame identifier, and sequentially stores the lossless residual data in an on-board cache area; an entropy tendency prediction module, which specifically extracts the motion vector field generated by the constant-rate lossy encoding module in real time, and calculates the statistical parameters of the motion vector field to generate a prediction signal representing the future information entropy tendency; an asynchronous compensation module, which specifically receives the prediction signal, calculates the difference between the constant target code rate and the actual encoding code rate of the constant-rate lossy encoding module to determine the code rate surplus, and adjusts the rate of taking out the lossless residual data from the head of the on-board cache area according to the prediction signal, and transmits the taken-out lossless residual data using the channel bandwidth corresponding to the code rate surplus.

2. The unmanned aerial vehicle data acquisition and transmission system based on lossless compression technology according to claim 1, characterized in that, The asynchronous compensation module further specifically encapsulates the taken-out lossless residual data into a compensation data packet carrying the frame identifier associated therewith, and embeds the compensation data packet into the lossy video stream being transmitted, so as to be synchronously sent in a single data channel.

3. The unmanned aerial vehicle data acquisition and transmission system based on lossless compression technology according to claim 1, characterized in that, The entropy potential prediction module calculates statistical parameters of the motion vector field, specifically, calculates the variance of all motion vector angles in the motion vector field , the calculation rule is , wherein, is the total number of motion vectors of the current frame, is the angle of the th motion vector, is the average of all motion vector angles; and when the value of the variance is lower than the first threshold value, the prediction signal instructs the asynchronous compensation module to reduce the retrieval rate of lossless residual data.

4. The unmanned aerial vehicle data acquisition and transmission system based on lossless compression technology according to claim 1, characterized in that, The system further comprises a ground-end fusion reconstruction module, which specifically receives and separates the lossy video stream and the compensation data packet from the transmission stream, decodes the lossy video stream, and superimposes the lossless residual data and the decoded lossy video frame of the corresponding frame identifier at the pixel level according to the frame identifier carried by the compensation data packet, so as to restore the lossy video frame to a lossless video frame.

5. The unmanned aerial vehicle data acquisition and transmission system based on lossless compression technology according to claim 1, characterized in that, The system further comprises an adaptive region of interest generation module, which specifically calculates the energy values of all pixel macroblocks in the lossless residual data in a single video frame to generate a residual energy map, calculates the global average energy value of the residual energy map, and defines the coordinates of the connected region with an energy value higher than the global average energy value multiplied by a coefficient in the residual energy map as region of interest metadata, and transmits the region of interest metadata. The lossless residual generation and caching module further specifically decomposes the lossless residual data of each frame into a high-priority structural residual part containing low-frequency coefficients and a low-priority textural residual part containing high-frequency coefficients through integer discrete cosine transform and two-stage quantization processing, and the system further comprises a cache manager, which specifically discards the corresponding low-priority textural residual part from the lossless residual data stored in the on-board cache area when the occupancy rate of the on-board cache area exceeds a cache pressure threshold.

6. The unmanned aerial vehicle data acquisition and transmission system based on lossless compression technology according to claim 1, wherein, The system further comprises an image quality evaluation and priority calibration module, which specifically monitors the sequence of quantization parameter QP values inside the constant-rate lossy encoding module, and judges the corresponding original video frame as a low-quality frame when it is monitored that the drop amplitude of the quantization parameter QP value exceeds a QP change threshold in a single time window.

7. The unmanned aerial vehicle data acquisition and transmission system based on lossless compression technology according to claim 1, characterized in that, ​ And for lossless residual data generated by the low quality frame, a low recovery priority is marked when stored in the on-board cache area, so that the asynchronous compensation module preferentially transmits lossless residual data with a high recovery priority.

8. The unmanned aerial vehicle data acquisition and transmission system based on lossless compression technology according to claim 1, wherein, The system further comprises a pre-filtering module, which is located before the original video frame sequence enters the constant code rate lossy encoding module and the lossless residual generation and caching module, and specifically: according to a spectrum data recording the vibration frequency characteristics of the unmanned aerial vehicle; performing fast Fourier transform on the original video data to convert to the frequency domain; and in the frequency domain data, the energy of the frequency band corresponding to the vibration frequency characteristics is suppressed through a digital notch filter, and then the inverse fast Fourier transform is performed to restore the purified spatial domain image data.

9. The unmanned aerial vehicle data acquisition and transmission system based on lossless compression technology according to claim 1, wherein, The system further comprises a navigation complexity map generation module, which specifically: obtains the real-time occupancy rate of the on-board cache area at a fixed time frequency; reads the geographic position coordinates of the on-board global positioning system module at the same time; and associates the real-time occupancy rate with the geographic position coordinates to generate map data representing the information complexity of different geographic positions on the flight path.

10. The unmanned aerial vehicle data acquisition and transmission system based on lossless compression technology according to claim 1, wherein, The on-board cache area in the lossless residual generation and caching module is a circular cache based on random access memory, which follows the first-in-first-out data reading and writing rule.

Images