Target information continuous generation method and system

Abstract

The application provides a method and system for continuous generation of target information. The method comprises: constructing a detection model for detecting a target to be detected; acquiring a sequence remote sensing image; determining a first image in which a suspected target to be detected first appears from the sequence remote sensing image, detecting a first target in the first image through the detection model; extracting a plurality of second images in which the suspected target to be detected appears continuously after the first image from the sequence remote sensing image, detecting a plurality of second targets in the plurality of second images through the detection model to form a second target set; calculating a degree of suspicion of each second target in the second target set and the first target, screening part of the second targets from the plurality of second targets according to the degree of suspicion to form a high-suspicion target chain; performing neighborhood constraint processing on a second image corresponding to each second target in the high-suspicion target chain to obtain a plurality of third images; and counting the number of times that the second target appears in the plurality of third images to determine whether the first target is the target to be detected.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, more particularly, to a target information continuous generation method and system. BACKGROUND

[0002] With the development of computer technology and remote sensing technology, the spatial resolution and time resolution of remote sensing images have been greatly improved, providing more abundant information for image analysis using remote sensing images and time series, and the target information continuous generation technology based on sequence remote sensing images has a wide application in the field of traffic management.

[0003] Target detection based on image sequences is a hot direction of development in the field of future marine monitoring, and it is also an advantage field of target observation using video visible light remote sensing satellites. In the civil field, using remote sensing technology and computer vision and other technical means, the shape, texture, color and other characteristics in the remote sensing image can be analyzed, the background and target can be distinguished, and useful target information can be extracted to serve the application in traffic detection, environmental monitoring, marine detection, topographic mapping and resource exploration.

[0004] There is also target detection through single-frame image and trajectory extraction technology, and the detection speed of these two technologies is fast, but the multi-frame time sequence information of the image is not fully utilized, and there is a problem of high false alarm rate and missed alarm rate, which still has a large space for improvement in actual application.

[0005] In addition, optical remote sensing images are easily affected by clouds, fog, light and other factors, and the pixels of targets such as ships in the image are very few, which belong to small weak targets. Therefore, the difficulty of target detection and tracking of optical remote sensing sequence images mainly lies in that when the target motion speed changes or the wind direction changes, the characteristics of small weak targets change, and problems such as missed tracking and wrong tracking are easily caused.

[0006] The existing orbiting satellites can shoot high time resolution remote sensing image sequences or even videos. For example, the panchromatic light channel of Gaofen-4 satellite can shoot video-like image sequences with a frame rate of 5s at the fastest. At present, multi-frame correlation technology is mostly applied in natural image target tracking algorithms, and the research on ship target detection using multi-frame correlation in remote sensing images is still less. Therefore, it is feasible to use continuous multi-frame images for target detection to improve performance. SUMMARY

[0007] In view of the above problems, the present application provides a target information continuous generation method and system.

[0008] The application provides a target information continuous generation method, which comprises the following steps: constructing a detection model for detecting a target to be detected; acquiring a sequence remote sensing image to be processed, wherein the sequence remote sensing image comprises a plurality of continuous images; determining a first image in which a suspected target to be detected appears for the first time from the sequence remote sensing image, detecting a first target in the first image through the detection model; extracting a plurality of second images in which the suspected target to be detected appears continuously after the first image from the sequence remote sensing image, detecting a plurality of second targets in the plurality of second images through the detection model, and forming a second target set; calculating a suspected degree of each second target in the second target set and the first target, screening part of the second targets from the plurality of second targets according to the suspected degree, and forming a high-suspected target chain; performing neighborhood constraint processing on a second image corresponding to each second target in the high-suspected target chain, and obtaining a plurality of third images; and counting the number of times that the second target appears in the plurality of third images, and determining whether the first target is the target to be detected.

[0009] According to the embodiment of the application, the YOLO model is selected, the SENet channel attention mechanism is used to optimize the YOLO model, the BiFPN module is added to the optimized YOLO model, the convolution layer in the YOLO model after the addition is replaced by the fractal convolution module, and the detection model is formed.

[0010] According to the embodiment of the application, the time sequence correlation measurement value of the target to be detected is obtained from the detection model; and the features of the plurality of second targets in the plurality of second images are extracted through the time sequence correlation measurement value.

[0011] According to the embodiment of the application, the sequence signal is sent, the sequence signal sent by the second target of the previous frame is associated with the sequence signal sent by the second target of the next frame; for the plurality of second targets, the sequence signal of the previous second target is associated with the sequence signal of the next second target to form an associated sequence; and the plurality of associated sequences are determined as the second target set.

[0012] According to the embodiment of the application, the quantification representation model is established through the distribution rule of the features of the second target; and the suspected degree between each associated sequence in the set and the time sequence correlation measurement value is determined through the quantification representation model.

[0013] According to the embodiment of the application, when the sequence signals of any two second targets in the plurality of second targets are the same, the sequence signals of the any two second targets are associated to form an associated sequence; the associated sequence signals are marked as a sub-chain under a same second target mother chain; the plurality of sub-chains are merged and judged to obtain sub-chains with different suspected degrees, and a high-suspected target chain is formed.

[0014] According to the embodiment of the present application, when the plurality of sub-chains satisfy the specified condition, the sub-chain with the highest suspected degree is reserved; when the plurality of sub-chains do not satisfy the specified condition, the sub-chain is separated to become a new second target parent chain.

[0015] According to the embodiment of the present application, the high-suspected-degree target chain is subjected to data processing to suppress the complex background of the high-suspected-degree target chain; the processed high-suspected-degree target chain is subjected to filtering processing to remove clutter of the complex background; the high-suspected-degree target chain after removal of the clutter is subjected to neighborhood convolution processing to obtain information of the second target; each two adjacent second images are calculated through the information of the second target to obtain a plurality of track paths of the second target; the plurality of track paths of the second target are subjected to constraint processing to obtain a plurality of third images.

[0016] According to the embodiment of the present application, the plurality of third images are subjected to searching to detect whether the second target exists in each third image, wherein when the track path of the second target satisfies the constraint condition, the third image has the second target; the number of times of appearance of the second target in the plurality of third images is counted and compared with a set threshold value; when the number of times of appearance is greater than or equal to the threshold value, the first target is the target to be detected.

[0017] Another aspect of the present application provides a target information continuous generation system, comprising: a target detection module, configured to collect characteristics of a target to be detected, and construct a detection model for detecting the target to be detected; a target extraction module, configured to acquire a sequence remote sensing image to be processed, the sequence remote sensing image comprising a plurality of continuous images, determine a first image in which a suspected target to be detected first appears from the sequence remote sensing image, and extract a plurality of second images from the sequence remote sensing image, the plurality of second images being continuous after the first image; a target processing module, configured to perform suspected degree calculation and neighborhood constraint processing on the plurality of second images to obtain a plurality of third images; and a target judgment module, configured to perform threshold value comparison on the plurality of third images to confirm whether a first target in which the suspected target to be detected first appears in the first image is the target to be detected.

[0018] The target information continuous generation method and system provided by the present application can achieve the following beneficial effects:

[0019] Considering the high-resolution characteristics of the visible light image, a kind of infrared small target fast detection algorithm based on background difference method is designed for pure sea or pure cloud background in infrared image, and combined with the motion characteristics of aircraft target, a kind of interframe target association algorithm based on multi-frame target mapping is used to extract target track and suppress false alarm, solve the problems of invalid feature matching, missing target, missing tracking and wrong tracking, effectively reduce the false alarm rate and the missing alarm rate, and improve the tracking speed and robustness in complex background. BRIEF DESCRIPTION OF DRAWINGS

[0020] The above and other objects, features and advantages of the present application will become more apparent from the following description of embodiments of the present application taken in conjunction with the accompanying drawings, in which:

[0021] Figure 1 A flow chart of a target information continuous generation method according to an embodiment of the present application is schematically shown;

[0022] Figure 2 A principle diagram of target information image processing according to an embodiment of the present application is schematically shown;

[0023] Figure 3 A flow chart of constructing a detection model for detecting a target to be detected according to an embodiment of the present application is schematically shown;

[0024] Figure 4 A structure diagram of a Dense-Yolo network model structure according to an embodiment of the present application is schematically shown.

[0025] Figure 5 A principle diagram of fractal convolution module and conventional convolution layer feature collection according to an embodiment of the present application is schematically shown.

[0026] Figure 6 A principle diagram of forming a target association sequence according to an embodiment of the present application is schematically shown;

[0027] Figure 7 A principle diagram of merging judgment of a sub-chain in a target chain according to an embodiment of the present application is schematically shown;

[0028] Figure 8 A flow chart of processing a second image to obtain a third image according to an embodiment of the present application is schematically shown;

[0029] Figure 9 A block diagram of a target information continuous generation system according to an embodiment of the present application is schematically shown.

[0030] Explanation of reference numerals:

[0031] 100 - first image; 200 - second image; 300 - third image; 400 - sequence signal; 500 - computer; 600 - mother chain; 601 - No. 1 sub-chain; 602 - No. 2 sub-chain; 603 - No. 3 sub-chain; 604 - No. 4 sub-chain. DETAILED DESCRIPTION

[0032] Embodiments of the present application will be described herein below with reference to the accompanying drawings. It is to be understood, however, that the description is merely exemplary and is not intended to limit the scope of the present application. In the following detailed description of embodiments of the present application, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that one or more embodiments of the present application can be practiced without these specific details. In other instances, well-known structures and techniques have not been described in detail in order to avoid obscuring aspects of the present application.

[0033] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used herein, the term "includes" and tautological expressions thereof, such as "including," "includes," "include," "contains," "containing," and so forth, mean the term "comprises."

[0034] All terms used herein, including technical and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined herein. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of the specification, and should not be interpreted in an idealized or overly formal manner.

[0035] Before describing specific embodiments of the present application in detail, technical terms are first explained to facilitate better understanding of the present application.

[0036] Sequence remote sensing imagery: refers to multiple frames of images acquired by satellites, unmanned aerial vehicles or other remote sensing platforms at consecutive time points, which have relevance in time, space and spectral dimensions, and can be used for dynamic monitoring, target tracking and change detection tasks.

[0037] Neighborhood constraint: refers to the use of the relevance of the local area (neighborhood) around the target object in data processing or model optimization to impose restrictions or guiding conditions on the results, in order to improve accuracy, continuity or physical reasonableness, so that the properties or behaviors of the target object should maintain a certain relevance with its neighboring objects.

[0038] In view of this, the present application provides a target information continuous generation method and system.

[0039] Figure 1 A flowchart of a target information continuous generation method according to an embodiment of the present application is schematically shown; Figure 2 A schematic diagram of the principle of target information image processing according to an embodiment of the present application is shown.

[0040] As Figure 1 and Figure 2 The target information continuous generation method according to this embodiment includes steps S1-S7.

[0041] At step S1, a detection model for detecting a to-be-detected target is constructed.

[0042] For example, the to-be-detected target can be an airplane. Different models of airplanes have different sizes, and a detection model based on deep learning of optical remote sensing images is constructed by combining local environment information perception, difficult example sample mining, BiFPN multi-scale fusion strategy, and small target guarantee mechanisms such as deconvolution, dense connection, and multi-scale receptive field mechanism.

[0043] For example, the detection model can use a YOLO model, a Faster R-CNN, etc.

[0044] At step S2, a sequence remote sensing image to be processed is obtained, and the sequence remote sensing image includes continuous multiple images.

[0045] At step S3, a first image 100 in which a suspected to-be-detected target first appears is determined from the sequence remote sensing image, and a first target in the first image 100 is detected by using the detection model.

[0046] The first image 100 in which the suspected to-be-detected target first appears is located from the sequence remote sensing image, is quickly screened by using a time series difference or an anomaly detection algorithm, is analyzed by using the detection model, and first target information is recognized and located, and the information includes an output target category, a position, and a confidence, etc.

[0047] At step S4, multiple second images 200 in which suspected to-be-detected targets appear after the first image 100 are extracted from the sequence remote sensing image, multiple second targets in the multiple second images 200 are detected by using the detection model, and a second target set is formed.

[0048] The multiple second images 200 in which the suspected to-be-detected targets appear after the first image 100 are extracted from the sequence remote sensing image, time continuity and spatial coverage consistency are ensured, the multiple second targets in the second images 200 are analyzed frame by frame, target positions, categories, and confidences are output, and the second target set is formed.

[0049] At step S5, a suspected degree of each second target in the second target set and the first target is calculated, a part of the second targets is screened from the multiple second targets according to the suspected degree, and a high-suspected-degree target chain is formed.

[0050] Features such as appearances, textures, spectral features, and motion vectors of the first target and each target in the second target set are extracted, the suspected degree is calculated by similarity measurement, a target motion model such as Kalman filter prediction position or a time series context such as optical flow trajectory consistency is combined to adjust the suspected degree weight, the second targets with a suspected degree higher than a set value are reserved, and the high-suspected-degree targets that are overlapped in space or time series are removed to avoid redundancy.

[0051] For example, the similarity measure can employ one or more of cosine similarity, IoU, HOG feature matching.

[0052] In step S6, the neighborhood constraint processing is performed on the second image 200 corresponding to each second target in the high-similarity target chain, to obtain a plurality of third images 300.

[0053] A peripheral region of a fixed range or an adaptive range (based on the target size) is extracted with each high-suspected target as the center, the neighborhood boundary is dynamically adjusted in combination with the target positions of the front and rear frames, the pixel labels in the neighborhood are optimized, the smooth transition of the target and the background is ensured, the target features in the neighborhood are strengthened, the guided filtering or wavelet transform is used to suppress noise, the regions in the neighborhood are processed, the other parts of the image remain unchanged, and the plurality of third images 300 are output, so as to improve the saliency of the target in the complex background.

[0054] In step S7, the number of times that the second target appears in the plurality of third images 300 is counted, and it is determined whether the first target is the target to be detected.

[0055] In the detection process, the image in which the suspected target to be detected first appears is taken as an initial image, and the position of the first target is taken as an initial position. The next N detection result images are searched, and for each image detection, if the second target is detected, the number of times obtained after all the second targets are accumulated is greater than or equal to a set threshold d, it is considered that the first target in the initial image is the real target to be detected, and if it is less than the threshold d, the first target is excluded.

[0056] Figure 3 A flowchart for constructing a detection model according to an embodiment of the present application is schematically shown; Figure 4 A structural diagram of a Dense-Yolo network model structure according to an embodiment of the present application is schematically shown; Figure 5 A principle diagram of fractal convolution module and conventional convolution layer feature collection according to an embodiment of the present application is schematically shown.

[0057] As shown in Figure 3 , Figure 4 and Figure 5 , the detection model for detecting the target to be detected according to the embodiment is constructed, including steps S11-S12.

[0058] In step S11, the YOLO model is selected, and the YOLO model is optimized by using the SE-Net channel attention mechanism.

[0059] Aiming at the characteristics of different sizes of different types of aircraft and the application requirements of target rapid detection, the time-sensitive YOLO network is used as the basis, and targeted improvements are made to propose a deep learning target detection algorithm with better scale adaptability-Dense-Yolo. The Dense-Yolo network model structure adopts DenseNet network, and based on the network dense connection idea of DenseNet network, the residual network structure Resunit in the two CSP1_3 modules in the backbone network is replaced with the custom Denseblock module, and the squeeze and excitation method in SENet is used to optimize the network structure, so as to explicitly model the interdependence between feature channels.

[0060] For example, the difference between DenseNet network and residual network is that the output of the Nth layer of Resunit module comes from the output of the (N-1)th layer plus a nonlinear transformation of the output of the (N-1)th layer, and the specific formula is as follows:

[0061]

[0062] In the formula, y N represents the output of the Nth layer, f N represents the nonlinear transformation of the Nth layer, and x N-1 represents the output of the (N-1)th layer.

[0063] For example, the specific formula of Denseblock module is as follows:

[0064]

[0065] In the formula, y N represents the output of the Nth layer, f N represents the nonlinear transformation of the Nth layer, and x N-1 represents the output of the (N-1)th layer.

[0066] In step S12, the BiFPN module is added to the optimized YOLO model, the convolution layer in the YOLO model after adding is replaced with the fractal convolution module, and the detection model is formed.

[0067] ​​​​​​​​​​BiFPN module is added to improve the scale compatibility of small-scale targets. FPN is an enhancement of the traditional CNN network for expressing and outputting image information, aiming to improve the feature extraction method of the CNN network, so that the final output features can better represent the information of each dimension of the input image. BiFPN adds a weight weighting mechanism on the basis of the complex bidirectional fusion of the FPN network, that is, different scales are given a weight value. The traditional method is to directly stack features of different scales, while BiFPN allows the network to learn the weight of different input features, and more efficiently bidirectionally fuses features of different scales.

[0068] The fractal convolution module has the ability to improve the prediction and regression of different types of aircraft targets with multiple width-height ratios. When using the conventional convolution layer in the YOLO model to collect features, the features collected on the narrow side are relatively sparse, which will lead to insufficient feature information for aircraft type prediction and narrow side bounding box regression, making it difficult to support good prediction and regression. Unlike conventional convolution layers, the fractal convolution module ensures consistent feature sampling rate for feature collection in the width and height dimensions of the aircraft as much as possible, ensuring reasonable collection of aircraft target key point feature information.

[0069] For example, the regression layer in the fractal convolution module accelerates the convergence speed of the network to a certain extent. The convergence speed formula is expressed as:

[0070]

[0071]

[0072] In the formula, represents the convergence speed in the width direction, represents the convergence speed in the height direction, represents the convolution step size on the i-th feature layer, represents the prediction value in the width direction on the i-th feature layer, represents the prediction value in the height direction on the i-th feature layer, represents the actual value in the width direction, represents the prediction value in the width direction, represents the actual value in the height direction, represents the prediction value in the height direction.

[0073] Figure 6 The principle diagram of forming a target association sequence according to an embodiment of the application is schematically shown.

[0074] As Figure 6As shown, the sequence signal 400 is sent by the computer 500, each second target sends a unique sequence signal 400 in a continuous frame, the sequence signal 400 includes metadata such as target name, position, speed, confidence, etc., the sequence signal 400 sent by the second target in the previous frame is associated with the sequence signal 400 sent by the second target in the next frame, for multiple second targets, the sequence signal 400 of the previous second target is associated with the sequence signal 400 of the next second target, for each second target, the cross-frame sequence signal 400 thereof is spliced in chronological order to form multiple associated sequences, which constitute a second target set.

[0075] For example, the pre-sequenced sequence can filter and associate the second target of the subsequent frame according to the characteristics of the finiteness of the speed, the energy, the stability of the speed direction and the size change, or can associate the second target after a certain interval frame number, and update the data of the sequence with the associated point.

[0076] Figure 7 The schematic diagram of the merging judgment of the sub-chain in the target chain according to the embodiment of the application is shown.

[0077] As shown in Figure 7 When the sequence signals 400 of any two second targets in the multiple second targets are the same, the sequence signals 400 of the any two second targets are associated to form an associated sequence, and the associated sequence signal 400 is marked as a sub-chain under the same second target mother chain 600. The multiple sub-chains (No. 1 sub-chain 601, No. 2 sub-chain 602, No. 3 sub-chain 603, and No. 4 sub-chain 604) are subjected to merging judgment to obtain sub-chains with different degrees of suspicion, form a high-suspected target chain, when the multiple sub-chains (No. 1 sub-chain 601, No. 2 sub-chain 602, No. 3 sub-chain 603, and No. 4 sub-chain 604) meet the specified conditions, the sub-chain with the highest degree of suspicion is retained, and when the multiple sub-chains do not meet the specified conditions, the sub-chain is separated to become a new second target mother chain 600.

[0078] For example, the sequence is subjected to an interruption threshold value and an “M selects N” condition judgment, if the sequence meets the discard condition corresponding to any one of the above judgments, the sequence is discarded, a quantitative representation model is established by establishing the feature distribution law of the second target in different dimensions (displacement, speed, energy, distance between the measured position and the predicted position, etc.), and the similarity degree of the features of the suspected target associated sequence relative to the quantitative values of the second target features is measured.

[0079] For example, when the same sequence is associated with multiple second targets, the new sequences after association are identified as different sub-chains (sub-chain 1 601, sub-chain 2 602, sub-chain 3 603, sub-chain 4 604) under the same parent chain 600. Sub-chains under the same parent chain 600 that subsequently converge at the same point are merged. If the conditions are met, only the sub-chain with the highest likelihood (sub-chain 3 603) is retained, and the remaining sub-chains (sub-chain 1 601, sub-chain 2 602, sub-chain 4 604) are discarded. If the conditions are not met and the generated sub-chain frame number is greater than the current frame number than the tolerance threshold, then the sub-chains (sub-chain 1 601, sub-chain 2 602, sub-chain 3 603, sub-chain 4 604) are separated to become new independent parent chains 600, with three possible forms:

[0080] (a) Subchain 1, 601, becomes the new parent chain 600. Subchain 2, 602, 603, and 604 are all subchains under this new parent chain 600.

[0081] (b) Subchain 1, 601, becomes the new parent chain 600; subchain 2, 602, becomes the subchain of subchain 1, 601; subchain 3, 603, becomes the new parent chain 600; and subchain 4, 604, becomes the subchain of subchain 3, 603.

[0082] (c) Subchain 1 (601), subchain 2 (602), subchain 3 (603), and subchain 4 (604) are all used as the new parent chain 600.

[0083] Figure 8 The flowchart illustrating the processing of a second image 200 to obtain a third image 300 according to an embodiment of the present invention is shown.

[0084] like Figure 8 As shown, processing the second image 200 to obtain the third image 300 according to an embodiment of the present invention includes steps S61 to S64.

[0085] In step S61, data processing is performed on the high-probability target chain to suppress the complex background of the high-probability target chain.

[0086] In step S62, the processed high-probability target chain is filtered to remove clutter from the complex background.

[0087] By applying minimum relative gray-level gradient filtering to the bidirectional oversampled image of the target, noise can be effectively reduced. The purpose of minimum relative gray-level gradient filtering is to filter out high-intensity noise and drastically changing background clutter.

[0088] For example, the filtering process can also employ: (1) guided filtering, using a guide image (such as a target neighborhood) to constrain the filtering process and preserve edges; (2) bilateral filtering, combining spatial distance and pixel value similarity for weighted smoothing; (3) non-local mean: using global similar blocks of the image for weighted averaging.

[0089] In step S63, the neighborhood convolution processing is performed on the high-suspected target chain after the clutter removal, and the information of the second target is obtained.

[0090] In the noise reduction processing of the second image 200, the neighborhood convolution processing is performed on the second target, and the convolution result is used to replace the pixel value of the target. The more bright pixels around the second target, the higher the brightness gain after the convolution. The determination of the convolution window makes the bright pixels closer to the center of the window have a greater contribution gain, which helps to preserve the information of the second target.

[0091] In step S64, the second target information is used to calculate each two adjacent second images 200, and the tracks of multiple second targets are obtained. The tracks of the multiple second targets are constrained, and the multiple third images 300 are obtained.

[0092] In the infrared image sequence, the design of the arrangement interval makes the distance between the target points formed by the second target in the sky in the difference image of two adjacent images in the sequence and the background meet the neighborhood constraint condition. The distance of the same target on adjacent frames is related to the speed range of the target.

[0093] For example, the target similarity (r) is represented as the correlation coefficient of the small area of the detected target in the two sequence images, and the formula is represented as

[0094]

[0095] In the formula, represents the mean value of the area A, represents the mean value of the area B, represents the observation value of the coordinates (x, y) in the area A, represents the observation value of the coordinates (x, y) in the area B.

[0096] In order to avoid the influence of the target on the calculation results of the mean value and the mean square deviation, the center area of the neighborhood image is first removed, and the mean value and the variance are calculated only for the area excluding the target.

[0097] Based on the method disclosed in the above embodiment, the application also provides a target information continuous generation system, which will be described in detail below. Figure 9 The system will be described in detail. ​​​​

[0098] Figure 9 A block diagram of a target information continuous generation system according to an embodiment of the present application is shown schematically.

[0099] As shown in Figure 9 the target information continuous generation system 700 according to the embodiment includes a target detection module 710, a target extraction module 720, a target processing module 730, and a target judgment module 740.

[0100] The target detection module 710 is configured to collect features of a target to be detected, and construct a detection model for detecting the target to be detected.

[0101] The target extraction module 720 is configured to acquire a sequence remote sensing image to be processed, the sequence remote sensing image including a plurality of continuous images, determine a first image 100 in which a suspected target to be detected first appears from the sequence remote sensing image, and extract a plurality of second images 200 from the sequence remote sensing image, the plurality of second images 200 being continuous after the first image 100.

[0102] The target processing module 730 is configured to perform suspected degree calculation and neighborhood constraint processing on the plurality of second images 200 to obtain a plurality of third images 300.

[0103] The target judgment module 740 is configured to perform threshold comparison on the plurality of third images 300 to determine whether a first target in which the suspected target to be detected first appears in the first image 100 is the target to be detected.

[0104] It should be noted that the embodiment of the system part is similar to the embodiment of the method part described above, and the technical effects achieved are also similar. For specific details, please refer to the method embodiment part described above, which will not be repeated here.

[0105] According to embodiments of the present application, any of the target detection module 710, the target extraction module 720, the target processing module 730, and the target judgment module 740 can be combined in one module, or any of the modules can be split into multiple modules. Alternatively, at least part of the function of one or more of the modules can be combined with at least part of the function of the other modules, and implemented in one module. According to embodiments of the present application, at least one of the target detection module 710, the target extraction module 720, the target processing module 730, and the target judgment module 740 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system on chip, a system on board, a system in package, an application specific integrated circuit (ASIC), or any other reasonable manner of integrating or packaging circuitry, etc., or implemented in hardware or firmware, or implemented in any one of software, hardware, and firmware, or implemented in a proper combination of any of them. Alternatively, at least one of the target detection module 710, the target extraction module 720, the target processing module 730, and the target judgment module 740 can be at least partially implemented as a computer program module that can perform the corresponding function when the computer program module is run.

[0106] The flow charts and block diagrams in the drawings are illustrations of the possible architectures, functionalities, and operations of apparatuses, methods, and systems according to various embodiments of the present application. In this regard, each block in the flow charts or block diagrams can represent a module, a segment, or a portion of code that comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or in the reverse order, depending on the functionality involved. It is also noted that each block in the block diagrams or flow charts, and combinations of blocks in the block diagrams or flow charts, can be implemented by dedicated hardware-based systems that perform the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0107] Those skilled in the art can understand that the features described in various embodiments of the present application can be combined or integrated in various combinations or integrations, even if such combinations or integrations are not explicitly described in the present application. In particular, the features described in various embodiments of the present application can be combined or integrated in various combinations or integrations without departing from the spirit and teachings of the present application. All such combinations or integrations fall within the scope of the present application.

[0108] The embodiments of the application have been described. However, these embodiments are merely for illustration and are not intended to limit the scope of the application. Although each embodiment is described above separately, this does not mean that the measures in each embodiment cannot be used advantageously in combination. Various alternatives and modifications to the embodiments described herein will be apparent to those skilled in the art in view of the foregoing without departing from the scope of the application.

Claims

1. A method of continuously generating target information, characterized by, The method comprises the following steps: constructing a detection model for detecting a target to be detected, comprising: selecting a YOLO model, and optimizing the YOLO model by using a SENet channel attention mechanism; adding a BiFPN module to the optimized YOLO model, replacing the convolutional layer in the YOLO model after the addition with a fractal convolutional module to form the detection model; obtaining a sequence remote sensing image to be processed, the sequence remote sensing image comprising a plurality of continuous images; determining a first image in which a suspected target to be detected first appears from the sequence remote sensing image, and detecting a first target in the first image by using the detection model; extracting a plurality of second images in which a suspected target to be detected appears after the first image from the sequence remote sensing image, detecting a plurality of second targets in the plurality of second images by using the detection model, and forming a second target set; wherein detecting a plurality of second targets in the plurality of second images by using the detection model comprises: obtaining a time sequence correlation measure value of the target to be detected from the detection model; extracting features of the plurality of second targets in the plurality of second images by using the time sequence correlation measure value; calculating a suspected degree of each second target in the second target set with respect to the first target, screening part of the second targets from the plurality of second targets according to the suspected degree, and forming a high suspected degree target chain; performing neighborhood constraint processing on each second image corresponding to each second target in the high suspected degree target chain to obtain a plurality of third images; counting the number of times that the second target appears in the plurality of third images, and determining whether the first target is the target to be detected.

2. The method of claim 1, wherein, The features of the second target comprise a sequence signal, and the forming of the second target set further comprises: sending the sequence signal, and correlating the sequence signal sent by a second target in a previous frame with the sequence signal sent by a second target in a next frame; correlating the sequence signal of a previous second target with the sequence signal of a next second target for the plurality of second targets to form an association sequence; determining a plurality of association sequences as the second target set.

3. The method of claim 2, wherein, The calculating of the suspected degree of each second target in the second target set with respect to the first target comprises: establishing a quantitative representation model by using the distribution rule of the features of the second target; determining the suspected degree between each association sequence in the set and the time sequence correlation measure value by using the quantitative representation model.

4. The method of claim 2, wherein, The screening of part of the second targets from the plurality of second targets according to the suspected degree to form a high suspected degree target chain, the target chain comprising a plurality of parent chains and at least one sub-chain belonging to each parent chain, further comprises: when the sequence signals of any two second targets in the plurality of second targets are the same, correlating the sequence signals of the any two second targets to form an association sequence; identifying the correlated sequence signals as sub-chains under the same second target parent chain; merging and judging a plurality of the sub-chains to obtain sub-chains of different suspected degrees, and forming a high suspected degree target chain.

5. The method of claim 4, wherein, The merging and judging of the sub-chains comprises: When multiple sub-chains satisfy the specified condition, the sub-chain with the highest suspicion degree is retained; When multiple sub-chains do not satisfy the specified condition, the sub-chain is separated out to become a new second target parent chain.

6. The method of claim 1, wherein, The second image corresponding to each second target in the high-suspicion-degree target chain is subjected to neighborhood constraint processing to obtain multiple third images, including: Data processing is performed on the high-suspicion-degree target chain to suppress the complex background of the high-suspicion-degree target chain; Filtering processing is performed on the processed high-suspicion-degree target chain to remove clutter of the complex background; Neighborhood convolution processing is performed on the high-suspicion-degree target chain after the clutter is removed to obtain information of the second target; Each two adjacent second images are calculated through the information of the second target to obtain multiple tracks of the second target; Constraint processing is performed on multiple tracks of the second target to obtain the multiple third images.

7. The method of claim 6, wherein, The number of times the second target appears in the multiple third images is counted to determine whether the first target is the target to be detected, including: The multiple third images are searched to detect whether the second target exists in each third image, When the track of the second target satisfies the constraint condition, the third image exists the second target; The number of times the second target appears in the multiple third images is counted and compared with a set threshold value; When greater than or equal to the threshold value, the first target is the target to be detected.

8. A target information continuous generation system applying the target information continuous generation method according to any one of claims 1 to 7, characterized by Including: A target detection module is configured to collect characteristics of a target to be detected and construct a detection model for detecting the target to be detected; A target extraction module is configured to acquire a sequence remote sensing image to be processed, the sequence remote sensing image including multiple continuous images, determine a first image in which a suspected target to be detected first appears from the sequence remote sensing image, and extract multiple second images from the sequence remote sensing image, the multiple second images being continuous after the first image; A target processing module is configured to perform suspicion degree calculation and neighborhood constraint processing on the multiple second images to obtain multiple third images; A target judgment module is configured to perform threshold value comparison on the multiple third images to determine whether a first target in which the suspected target to be detected first appears in the first image is the target to be detected.

Images