Search

CN-121998840-A - Unmanned aerial vehicle aerial photographing area image quality optimization method based on image processing

CN121998840ACN 121998840 ACN121998840 ACN 121998840ACN-121998840-A

Abstract

The invention relates to the technical field of intelligent processing of unmanned aerial vehicle aerial image, and discloses an unmanned aerial vehicle aerial image area image quality optimization method based on image processing. The method synchronously receives an original image stream and a flight state data stream, identifies and quantizes various degradation factors through perceptual degradation analysis, and generates a feature descriptor containing degradation components and weights. Based on the descriptor, an adaptive search topology network which takes image quality expectation as traction and can reflect quality evolution paths is dynamically constructed. In this network, a multi-level guided focused search is performed depending on the type of degradation and intensity, locking the most relevant target image units to form a cluster to be processed. And implementing a parameter self-adaptive mixed enhancement strategy for the cluster, correcting and compensating for different degradation components, generating an optimized image and feeding back to an updating network. The method realizes the accurate diagnosis and self-adaptive optimization processing of the mixed degradation of the aerial image.

Inventors

  • SHAO BIN
  • WANG XIN

Assignees

  • 宁波市前沿数字科技有限公司

Dates

Publication Date
20260508
Application Date
20260120

Claims (10)

  1. 1. The unmanned aerial vehicle aerial photographing area image quality optimization method based on image processing is characterized by comprising the following steps of: receiving an original image stream and a flight state data stream synchronously acquired by an unmanned plane platform in flight operation; Performing perceptual degradation analysis on the original image stream, identifying a plurality of degradation factors including atmospheric disturbance, imaging equipment inherent noise and geometric distortion caused by flight state, and quantifying the degradation degree of each degradation factor on the current image stream, thereby constructing a degradation characteristic descriptor comprising a plurality of degradation components and weights thereof; Dynamically constructing an adaptive search structure covering a target area and taking image quality expectation as traction based on the degradation characteristic descriptors, wherein the adaptive search structure organizes image units in an image library into a topological network capable of reflecting a quality evolution path; in the self-adaptive searching structure, a multi-stage guided focusing search is executed according to the degradation type and strength indicated by the degradation characteristic descriptor, a series of target image units most relevant to the degradation characteristic descriptor are screened and locked layer by layer, and a unit cluster to be processed is formed; And carrying out quality optimization processing on the unit cluster to be processed, wherein the quality optimization processing adopts a mixed enhancement strategy which carries out parameter self-adaption according to degradation characteristic descriptors, applies correction and compensation operations to different degradation components respectively, generates optimized image units, and feeds the optimized image units back to a self-adaption searching structure to update the network state of the optimized image units.
  2. 2. The method for optimizing image quality of an unmanned aerial vehicle (unmanned aerial vehicle) area based on image processing according to claim 1, wherein the performing perceptual degradation analysis on the original image stream, identifying a plurality of degradation factors including atmospheric disturbance, imaging device intrinsic noise and geometric distortion caused by flight state, and quantifying degradation degree of each degradation factor on a current image stream, and further constructing a degradation feature descriptor including a plurality of degradation components and weights thereof, comprises: extracting continuous frame sequences from an original image stream, and primarily separating dynamic degradation components and static degradation components by analyzing inter-frame differences and intra-frame statistical characteristics; carrying out space-time alignment and association mapping on the attitude angular speed and the position jitter information in the flight state data stream and the dynamic degradation components, and solving the specific parameters of three degradation factors, namely motion blur, geometric distortion and field of view instability caused by flight motion; Carrying out multi-scale texture analysis and noise power spectrum estimation on a single frame image in an original image stream, and separating an influence model of three degradation factors, namely fixed mode noise, random noise and optical attenuation caused by atmospheric scattering, which are determined by the characteristics of an imaging sensor; And carrying out cross verification on the preliminary result after the association of the flight state data stream and single frame analysis, quantifying the confidence coefficient and the influence weight of each degradation factor, and integrating the parameters of all the degradation factors, the influence model and the confidence coefficient and the weight thereof into a structured data object, wherein the structured data object is a degradation characteristic descriptor.
  3. 3. The method for optimizing image quality of an unmanned aerial vehicle (unmanned aerial vehicle) area based on image processing according to claim 2, wherein the dynamically constructing an adaptive search structure covering a target area and taking image quality as a traction based on degradation feature descriptors comprises: the method comprises the steps of pre-reading metadata and shallow visual characteristics of all image units in an aerial photographing region image library, and establishing a preliminary space-quality association mapping table according to acquisition time, geographic positions and basic quality scores of the image units; Taking the dominant degradation type in the degradation characteristic descriptor as query guidance, and screening out image units which are influenced by similar degradation problems in history and have obvious optimization processing effect from a space-quality association mapping table as initial seed nodes for constructing the self-adaptive search structure; Taking the initial seed node as a core, and establishing directed connection among nodes according to content continuity, quality inheritance relationship and time-space adjacency among image units, wherein the connection direction represents the direction in which quality optimization is possible to transmit or refer to, and the connection strength is determined by the similarity of characteristics among the nodes and the correlation of quality gradients; And introducing an online updating mechanism, wherein the online updating mechanism adjusts the connection strength and the topological relation of nodes in the self-adaptive search structure in real time according to the latest received original image stream and the corresponding degradation characteristic descriptors thereof, and for the newly-appearing degradation mode, the online updating mechanism creates a new node branch in the self-adaptive search structure to accommodate and characterize the degradation mode.
  4. 4. A method of optimizing image quality of an unmanned aerial vehicle (unmanned aerial vehicle) area based on image processing according to claim 3, wherein in the adaptive search structure, a multi-stage guided focused search is performed according to the degradation type and intensity indicated by the degradation feature descriptor, a series of target image units most relevant to the degradation feature descriptor are screened and locked layer by layer to form a cluster of units to be processed, and the method comprises: encoding the degradation feature descriptors into query vectors which can be understood by the adaptive search structure, and inputting the query vectors into an entry node of the adaptive search structure; Starting a first-stage search, calculating the matching degree between the query vector and the feature vector of the image unit represented by each neighborhood node in the direct neighborhood of the entry node, and selecting nodes with the matching degree exceeding a primary threshold value to form a primary candidate set; Starting a second-level search, taking each node in the primary candidate set as a new search starting point, exploring the downstream of the self-adaptive search structure, wherein the exploration depth is limited by a preset hop count, and in the downstream exploration process, not only calculating the direct matching degree of the query vector and the downstream node, but also evaluating the cumulative effect of the connection strength on the path from the primary candidate node to the downstream node, integrating the scores of the two nodes, and screening out a secondary candidate node; And performing de-duplication and merging on all secondary candidate nodes, sorting according to the comprehensive scores of the secondary candidate nodes, selecting a node set which is ranked at the front and meets a final threshold value, wherein the image units corresponding to the node set are locked into target image units, all the target image units jointly form a unit cluster to be processed, and simultaneously recording the complete retrieval path from the entry node to each target image unit.
  5. 5. The method for optimizing image quality of an unmanned aerial vehicle aerial photographing area based on image processing according to claim 4, wherein the performing quality optimization processing on the to-be-processed unit cluster, the quality optimization processing adopting a hybrid enhancement strategy that performs parameter adaptation according to degradation feature descriptors, comprises: analyzing the attribute of each target image unit in the unit cluster to be processed, and extracting retrieval path information associated with the target image unit from the self-adaptive search structure; Applying deconvolution filtering of specific direction and intensity to the target image unit according to parameters of the degradation component of motion blur in the degradation feature descriptor; performing atmospheric light curtain estimation and contrast recovery operations on the target image unit according to the influence model for the optical attenuation degradation component in the degradation feature descriptor; according to the model and the weight of various noise degradation components in the degradation characteristic descriptor, processing a target image unit by adopting a mode of combining adaptive filtering and wavelet domain threshold noise reduction; And fusing intermediate results after the correction and compensation operations for motion blur, geometric distortion, optical attenuation and various noises are respectively implemented, and distributing fusion coefficients according to weights of different degradation components in the fusion process to generate a final optimized image unit.
  6. 6. The method for optimizing image quality of an unmanned aerial vehicle (unmanned aerial vehicle) area based on image processing according to claim 5, wherein feeding back the optimized image unit into the adaptive search structure to update the network state thereof comprises: Calculating quality difference measurement of the optimized image unit and an original target image unit in a unit cluster to be processed, and absolute quality scores of the optimized image units; Creating a new node in the self-adaptive search structure for the optimized image unit according to the quality difference measurement and the absolute quality score, or updating the characteristic representation of the node corresponding to the original target image unit; According to the effectiveness of the optimization process and the effectiveness of the retrieval path, the intensity of all connections on the path from the entry node to the new node or the update node in the self-adaptive search structure is adjusted, the connection which leads to successful optimization is enhanced, and the invalid or inefficient connection is weakened; and archiving the newly generated optimized image unit and metadata thereof to an aerial region image library, and synchronously updating a space-quality association mapping table.
  7. 7. The unmanned aerial vehicle aerial photographing area image quality optimization method based on image processing according to claim 5, wherein the processing the target image unit by combining adaptive filtering and wavelet domain threshold denoising according to the model and the weight of each noise degradation component in the degradation characteristic descriptor comprises: Extracting noise model parameters aiming at fixed pattern noise and random noise from the degradation characteristic descriptors, and respectively influencing weights; Positioning the spatial distribution mode of the noise on the target image unit according to the model parameters of the fixed mode noise, and constructing a spatial variation filter matched with the spatial distribution mode, wherein the kernel function of the spatial variation filter is adaptively adjusted according to the local statistical characteristics of the noise mode so as to inhibit the fixed mode noise; determining the noise power spectrum characteristics of the random noise according to the model parameters of the random noise, calculating the optimal decomposition layer number for wavelet transformation based on the noise power spectrum characteristics, and setting different threshold functions for different sub-bands; Performing wavelet transformation on the image subjected to spatial mutation filtering to obtain high-frequency subband coefficients in multiple scales and directions; According to the threshold function, soft threshold or hard threshold processing is carried out on each high-frequency subband coefficient, the coefficient with the amplitude lower than the threshold value is regarded as noise to be suppressed, and the coefficient higher than the threshold value is reserved to maintain image details; performing wavelet inverse transformation on the high-frequency subband coefficient subjected to the threshold processing, and reconstructing an image from which random noise is removed; and according to the influence weights of the fixed pattern noise and the random noise extracted from the degradation characteristic descriptor, carrying out weighted fusion on the output result of the spatial mutation filtering and the output result of wavelet threshold noise reduction, and generating a final image after noise suppression.
  8. 8. The method for optimizing image quality of an unmanned aerial vehicle (unmanned aerial vehicle) area based on image processing according to claim 2, wherein the perceived degradation analysis further comprises an environmental context modeling step, comprising: collecting flight environment data synchronous with an original image stream, wherein the flight environment data at least comprises illumination intensity, atmospheric visibility and temperature and humidity; Establishing an experience correlation model between environment data and image degradation factors, and performing environment context calibration on degradation factor parameters calculated by image content analysis and flight state correlation by using the experience correlation model; the calibrated parameters are used as more accurate inputs for updating and correcting the degradation feature descriptors.
  9. 9. The method for optimizing the image quality of an unmanned aerial vehicle aerial photographing area based on image processing according to claim 3, wherein the dynamic construction process of the self-adaptive search structure further comprises a structure self-optimization link, comprising: periodically evaluating the retrieval efficiency and accuracy of the self-adaptive search structure in the history inquiry; When the retrieval efficiency or accuracy is lower than a set standard, triggering a structural self-optimization link, wherein the structural self-optimization link carries out iterative adjustment on node division rules, connection establishment criteria and connection strength calculation formulas of the self-adaptive search structure based on historical query logs and optimization result feedback; On the premise of ensuring compatibility with the existing data, the new rule after adjustment is applied to the self-adaptive search structure to finish the reconstruction of the topological structure and the parameter reconstruction.
  10. 10. The method for optimizing image quality of an unmanned aerial vehicle aerial photographing area based on image processing according to claim 4, wherein the multi-stage guided focusing search process further comprises a search path backtracking verification step, comprising: after the target image unit is locked and a unit cluster to be processed is formed, the system automatically performs reverse simulation verification along the recorded complete retrieval path; simulation verification proceeds by calculating whether the same target image unit is still stably retrieved along the same path assuming that the current degradation feature descriptor is known; If the success rate of the simulation verification is lower than a preset value, marking the retrieval path as an unstable path, and reducing the dependency weight on the retrieval path or the similar path in the follow-up retrieval.

Description

Unmanned aerial vehicle aerial photographing area image quality optimization method based on image processing Technical Field The invention relates to the technical field of intelligent processing of unmanned aerial vehicle aerial image, in particular to an unmanned aerial vehicle aerial image area image quality optimization method based on image processing. Background The existing unmanned aerial vehicle aerial image quality optimization method mainly depends on a preset general image enhancement algorithm or independent analysis of a single frame image. These techniques typically implement denoising, sharpening, or color enhancement based on pixel-domain statistics of the image or frequency-domain transforms. However, in dynamic flight operations, image degradation is caused by a variety of factors, such as atmospheric scattering, sensor noise, platform vibration, and motion blur. The existing method lacks collaborative resolution capability to multi-source isomerism degradation factors, and cannot effectively separate and quantify specific contributions of degradation components. The optimization algorithm is fixed in parameters or blind in adjustment, is difficult to adapt to complex and changeable imaging environments, and is unstable in processing effect. The prior art often employs a method of selecting reference data from an image library based on image content feature matching or global quality score ranking. Such retrieval mechanisms have weak relevance to the specific type and strength of degradation suffered by the image, failing to build an inherent quality correlation model between image library elements. The retrieval results are mostly single images with similar visual contents or higher quality scores, and clues or state references cannot be processed in a targeted manner for the current specific and mixed degradation modes, so that the follow-up optimization process lacks accurate guidance. The technical scheme is needed, the precise quantitative diagnosis of the hybrid degradation cause of the aerial image can be realized, and the processing basis which is most matched with the current degradation state can be dynamically retrieved from the historical data according to the diagnosis result, so that a self-adaptive and precise image quality optimization process is driven. Disclosure of Invention The invention aims to provide an unmanned aerial vehicle aerial photographing area image quality optimization method based on image processing so as to solve the problems in the background technology. In order to achieve the above object, the present invention provides an image processing-based unmanned aerial vehicle aerial photographing region image quality optimization method, which includes: receiving an original image stream and a flight state data stream synchronously acquired by an unmanned plane platform in flight operation; Performing perceptual degradation analysis on the original image stream, identifying a plurality of degradation factors including atmospheric disturbance, imaging equipment inherent noise and geometric distortion caused by flight state, and quantifying the degradation degree of each degradation factor on the current image stream, thereby constructing a degradation characteristic descriptor comprising a plurality of degradation components and weights thereof; Dynamically constructing an adaptive search structure covering a target area and taking image quality expectation as traction based on the degradation characteristic descriptors, wherein the adaptive search structure organizes image units in an image library into a topological network capable of reflecting a quality evolution path; in the self-adaptive searching structure, a multi-stage guided focusing search is executed according to the degradation type and strength indicated by the degradation characteristic descriptor, a series of target image units most relevant to the degradation characteristic descriptor are screened and locked layer by layer, and a unit cluster to be processed is formed; And carrying out quality optimization processing on the unit cluster to be processed, wherein the quality optimization processing adopts a mixed enhancement strategy which carries out parameter self-adaption according to degradation characteristic descriptors, applies correction and compensation operations to different degradation components respectively, generates optimized image units, and feeds the optimized image units back to a self-adaption searching structure to update the network state of the optimized image units. Preferably, the performing perceptual degradation analysis on the original image stream, identifying multiple degradation factors including atmospheric disturbance, imaging device intrinsic noise and geometric distortion caused by flight state, and quantifying degradation degree of each degradation factor on the current image stream, and further constructing a degradation feature descriptor including multip