Long-distance tiny target detection method and system

By acquiring the physical parameters and focal range of the sensor, combined with image segmentation and improved YOLOv8 network model, the problem of low detection accuracy and efficiency of long-distance small targets is solved, and efficient identification and detection of long-distance small targets is achieved.

CN119027664AActive Publication Date: 2024-11-26SHANDONG UNIV
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411130651.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-11-26
Estimated Expiration
2044-08-16

AI Technical Summary

Technical Problem

The prior art has limited recognition accuracy and efficiency in long-distance and small-scale object detection, especially YOLOv8 has insufficient detection capabilities for small-scale objects within a range of 200 meters and cannot meet the actual application needs.

Method used

By obtaining the diagonal dimensions, width and height ratios and resolution of the sensor, calculating the physical width and focal length range, combined with image segmentation methods and improved YOLOv8 network model (using MobileNetV3 backbone network), high resolution for long-distance small targets can be achieved with high resolution for long-distance small targets, using mobileNetV3 backbone network). Detect and identify.

Benefits of technology

It realizes continuous coverage from close to long distances 200 meters, obtains high-resolution images to be identified, improves the detection and recognition efficiency and accuracy of long-distance small targets, and adapts to real-time detection requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027664A_ABST
    Figure CN119027664A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of target detection, and provides a long-distance tiny target detection method and system, and the method comprises the steps: obtaining the diagonal size, the width-to-height ratio and the resolution of a sensor, calculating the physical width of the sensor based on the diagonal size and the width-to-height ratio of the sensor, and carrying out the detection of a tiny target based on the physical width and the resolution of the sensor. Calculating the physical size of each pixel, and converting the pixel width occupied by the target in the image into the physical width of the target; the actual width of the target and the distance range between the target and the sensor are obtained, and the focal length range is determined by combining the physical width of the target; and acquiring an image in the focal length range, obtaining a plurality of image blocks through an image segmentation method, detecting a target in each image block by adopting an image detection model, and integrating the targets in all the image blocks to obtain a detection result. And high-efficiency detection and identification of long-distance small targets are realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of target detection, and in particular relates to a method and system for detecting a long-distance tiny target. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the rapid development of computer vision technology, object detection has been widely used. However, existing technologies still face many challenges in detecting small objects at long distances. Traditional object detection methods have significant limitations in recognition accuracy and efficiency when facing long-distance and small objects. YOLOv8, as a fast and effective object detection framework, is widely used in real-time detection tasks.

[0004] However, YOLOv8 has shown shortcomings in long-distance and small target detection. First, although the original backbone network Darknet53 of YOLOv8 performs well in many scenarios, its complexity and computational complexity make it difficult to maintain high efficiency in real-time detection tasks. Second, the YOLOv8 algorithm has limited detection capabilities for small targets within a range of 200 meters and cannot meet the needs of practical applications. Summary of the invention

[0005] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a method and system for detecting small targets at a long distance, and proposes to determine the focal length range based on information such as the sensor diagonal size, aspect ratio, actual width of the resolution target, and distance range between the target and the sensor, so as to obtain a high-resolution image to be identified under the premise of a small target at a long distance, thereby realizing efficient detection and identification of small targets at a long distance.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A first aspect of the present invention provides a long-distance small target detection method, which comprises:

[0008] Obtain the sensor diagonal size, aspect ratio and resolution, calculate the physical width of the sensor based on the sensor diagonal size and aspect ratio, calculate the physical size of each pixel based on the sensor's physical width and resolution, and convert the pixel width of the target in the image into the physical width of the target;

[0009] Obtaining the actual width of the target and the distance range between the target and the sensor, and determining the focal length range in combination with the physical width of the target;

[0010] An image within the focal length range is acquired, and a plurality of image blocks are obtained through an image segmentation method. After the target in each image block is detected using an image detection model, the targets in all the image blocks are integrated to obtain a detection result.

[0011] Furthermore, the physical width of the sensor is:

[0012] W s =k×Q1

[0013] Among them, the scaling factor The resolution is P1×P2, the aspect ratio is Q1:Q2, and the diagonal size is H.

[0014] Furthermore, the physical width of the target is:

[0015]

[0016] Where, the physical width of the sensor is W s , the resolution is P1×P2, and W1 is the pixel width of the target in the image.

[0017] Furthermore, the relationship between the focal length range and the distance range between the target and the sensor is:

[0018]

[0019] Where f is the focal length, D is the distance between the lens and the target, W3 is the actual width of the target, and the physical width of the target is W2.

[0020] Furthermore, the image segmentation method is: segmenting the image into n1 parts in the long axis direction, segmenting the image into n2 parts in the short axis direction, and finally segmenting the image into n1+n2 image blocks.

[0021] Furthermore, the image detection model adopts YOLOv8, and the backbone network of YOLOv8 adopts MobileNetV3.

[0022] Furthermore, the data set used for training the image detection model is obtained by labeling images collected at various angles and heights after brightness enhancement, rotation and flipping.

[0023] A second aspect of the present invention provides a long-distance small target detection system, comprising:

[0024] The target physical width calculation module is configured to: obtain the diagonal size, aspect ratio and resolution of the sensor, calculate the physical width of the sensor based on the diagonal size and aspect ratio of the sensor, calculate the physical size of each pixel based on the physical width and resolution of the sensor, and convert the pixel width occupied by the target in the image into the physical width of the target;

[0025] A focal range determination module is configured to: obtain an actual width of the target and a distance range between the target and the sensor, and determine a focal range in combination with the physical width of the target;

[0026] The target detection module is configured to: acquire an image within the focal length range, obtain multiple image blocks through an image segmentation method, detect the target in each image block using an image detection model, and then integrate the targets in all image blocks to obtain a detection result.

[0027] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the long-distance small target detection method as described above.

[0028] The fourth aspect of the present invention provides a computer device, including a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, wherein when the processor executes the program, the steps in a long-distance small target detection method as described above are implemented.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] The present invention determines the focal range based on information such as the sensor's diagonal size, aspect ratio, actual width of the resolution target, and the distance range between the target and the sensor, so as to achieve continuous coverage from short distance to long distance of 200 meters, obtain high-resolution images to be identified, and further realize efficient detection and identification of small targets at long distances.

[0031] The present invention utilizes an image segmentation method to perform segmentation processing on a high-resolution image so as to accurately extract a target area.

[0032] In order to meet the needs of real-time detection, the present invention improves the backbone network of YOLOv8 and replaces its original Darknet53 backbone network with a lightweight MobileNetV3, thereby improving the detection speed and efficiency.

[0033] The present invention realizes efficient detection and recognition of small targets at long distances through the comprehensive application of network model optimization, multi-focal-segment shooting and image segmentation technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0035] Figure 1 It is a structural diagram of an improved YOLOv8 network model according to the first embodiment of the present invention;

[0036] Figure 2 It is a structural diagram of a computer device according to a fourth embodiment of the present invention. DETAILED DESCRIPTION

[0037] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0038] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0039] Embodiment 1

[0040] This embodiment provides a method for detecting small targets at a long distance.

[0041] This embodiment provides a long-distance tiny target detection method to solve the problem in the prior art that the recognition accuracy of small targets at long distances is low.

[0042] The present embodiment provides a method for detecting a small target at a long distance, comprising the following steps:

[0043] S1. Collect target images to form an initial data set, preprocess the initial data set to generate a processed data set, and use the LabelImg tool to label the data set.

[0044] Specifically, the initial data set of small targets is generated by collecting data at various angles and heights, and images with unclear target features or too messy features are deleted. The image resolution is 3840×2160; the initial data set is preprocessed by brightness enhancement, rotation, flipping, etc. to increase the diversity of the data set and improve the generalization ability and robustness of the model; the preprocessed data set is annotated using LabelImg and saved in the form of a txt file.

[0045] S2. Improve the YOLOv8 network model and replace its original Darknet53 backbone network with the lightweight MobileNetV3 to improve the detection speed and efficiency. Furthermore, train based on the improved YOLOv8 model to obtain the weight file.

[0046] Specifically, the original Darknet53 backbone network of YOLOv8 is replaced with the lightweight MobileNetV3, such as Figure 1 As shown in the figure, the lightweight model uses deep separable convolution to greatly reduce the amount of calculation and the number of parameters. The deep convolution applies convolution independently on each input channel, and then introduces mapping between different channels through point-by-point convolution. It is combined with the lightweight attention mechanism SE module to adaptively adjust the weight of each channel to enhance important features and suppress irrelevant information. In addition, it is combined with network structure search NAS. MobileNetV3 combines automatic architecture search technology to optimize the network structure to achieve a better balance between performance and efficiency. In general, it can improve detection speed and efficiency. Further, training is performed based on the improved lightweight network model to obtain the weight file best.pt.

[0047] S3. Set up 4 lenses with different focal lengths to achieve continuous coverage from short distance to long distance of 200 meters and obtain high-resolution images to be identified.

[0048] Step S3 specifically includes the following:

[0049] Step 301: Obtain the sensor diagonal size, aspect ratio, and resolution.

[0050] For example, when shooting with the IMX415 sensor, the diagonal size of the sensor is 6.43mm, the resolution is 3840×2160, and the aspect ratio is 16:9.

[0051] Step 302: Calculate the physical width and height of the sensor based on the diagonal size and the aspect ratio of the sensor.

[0052] First, calculate the scaling factor k:

[0053]

[0054] Therefore, the physical width Ws and height Hs of the sensor are:

[0055] W s =16×k=5.6mm

[0056] H s =9×k=3.15mm

[0057] Step 303: Calculate the physical size of each pixel based on the physical width and resolution of the sensor:

[0058]

[0059] Step 304: Convert the pixel width of the target in the image into the physical width:

[0060] W1×1.458×10 -3 mm / pixel=W2

[0061] Where W1 is the pixel width of the target in the image, in pixels; W2 is the physical width of the target, in mm.

[0062] In summary, the physical width of the target W s = k × Q1;

[0063] Among them, the resolution is P1×P2, P1 is the horizontal pixel, P2 is the vertical pixel; the aspect ratio is Q1:Q2; the diagonal size is H.

[0064] Step 305 : Determine a focal length range based on the physical width of the target, combined with the actual width of the target and the distance range between the target and the sensor.

[0065] Among them, the relationship between the focal length f and the distance range between the target and the sensor is:

[0066]

[0067] Where D is the distance between the lens and the target, in meters; W3 is the actual width of the target, in meters.

[0068] Considering small targets such as birds and bees, the ratio of the pixel width of the target in the image to the temporal width of the target is approximately between 280-550 pixel / m. The following is the range of lens focal length.

[0069] (1) The focal length range of the first lens is selected to be 21mm to 40mm. The lens within this focal length can provide a wide viewing angle, which is suitable for capturing a wide range of scenes. Therefore, this lens can shoot small targets within a close range (0-50 meters). Due to the wide viewing angle, the lens can ensure that the target occupies a larger pixel ratio in the image (the image resolution is 3840×2160), significantly improving the saliency and resolution of the target features, which helps the model to more accurately extract and identify small targets;

[0070] (2) The focal length range of the second lens is selected to be 41mm to 80mm. The lens angle of view and focal length within this focal length are moderate, which can take into account both the shooting range and details, and is suitable for shooting small targets within a medium distance (50-100 meters). Therefore, this lens can provide relatively clear target details within this distance range, while still retaining sufficient environmental background information to facilitate target positioning and identification. The focal length selection makes the target have a moderate pixel density in the image (the image resolution is 3840×2160), ensuring the optimization of feature extraction and recognition effects;

[0071] (3) The focal length range of the third lens is selected to be 63mm to 120mm. The lens within this focal length provides a narrow viewing angle, which is suitable for long-distance shooting. Therefore, this lens can shoot small targets within a long distance (101-150 meters). Using a lens with this focal length range can significantly increase the pixel ratio of the target in the image (the image resolution is 3840×2160), improve the resolution of the target features, and clearly capture the target details even at a long distance, thereby enhancing the detection accuracy;

[0072] (4) The focal length range of the fourth lens is selected to be 84mm to 160mm, which is suitable for extremely long-distance shooting and is dedicated to capturing small targets at a distance. Therefore, this lens can shoot small targets at extremely long distances (151-200 meters). The high focal length lens can significantly magnify small targets at a long distance so that they occupy sufficient pixel area in the image (the image resolution is 3840×2160), thereby enhancing the saliency and resolution of the target features, ensuring efficient and accurate recognition and detection of small targets at a long distance.

[0073] S4. Based on the small target photographed in step S3, the high-resolution image is segmented into multiple image blocks using an image segmentation method, and then input into an improved YOLOv8 network model (640×640) for detection.

[0074] Specifically, the picture taken in step S3 is segmented into multiple image blocks by using an image segmentation method. The picture is segmented into 4 parts with 960 pixels in the long axis direction (3840 pixels), and into 3 parts with 720 pixels in the short axis direction (2160 pixels). Finally, the image is segmented into 12 image blocks, and the size of each image block is 960×720, which is as close to the expected input size as possible although not 640×640. Each segmented image block is directly input into the improved YOLOv8 network model. The subsequent improved YOLOv8 model (image detection model) will process each image block to detect and identify small targets. The detection results of each image block are integrated and mapped back to the coordinate system of the original high-resolution image to ensure that each detected small target can be accurately located in the original full-frame image and maintain the integrity of the detection results.

[0075] The present embodiment provides a method for detecting small targets at a distance. First, the target image is collected to form an initial data set, and the data set is preprocessed to generate a processed data set. Subsequently, the LabelImg tool is used for labeling. In order to meet the real-time detection requirements, the backbone network of YOLOv8 is improved, and its original Darknet53 backbone network is replaced with a lightweight MobileNetV3, thereby improving the detection speed and efficiency, and training is performed on the basis of the improved YOLOv8 model. In view of the significant limitations of the YOLOv8 algorithm in the recognition of long-distance and small targets, it cannot meet the detection requirements of small targets at a long distance such as 200 meters. Therefore, lenses with different focal lengths are designed to achieve continuous coverage from close distance to a long distance of 200 meters, and obtain high-resolution images to be identified. In the case where small targets are difficult to effectively identify by direct image scaling, the high-resolution image is segmented by using an image segmentation method to accurately extract the target area. Through the comprehensive application of network model optimization, multi-focal-segment shooting and image segmentation technology, efficient detection and identification of small targets at a long distance is achieved.

[0076] Embodiment 2

[0077] This embodiment provides a long-distance small target detection system, which specifically includes:

[0078] The target physical width calculation module is configured to: obtain the diagonal size, aspect ratio and resolution of the sensor, calculate the physical width of the sensor based on the diagonal size and aspect ratio of the sensor, calculate the physical size of each pixel based on the physical width and resolution of the sensor, and convert the pixel width occupied by the target in the image into the physical width of the target;

[0079] A focal range determination module is configured to: obtain an actual width of the target and a distance range between the target and the sensor, and determine a focal range in combination with the physical width of the target;

[0080] The target detection module is configured to: acquire an image within the focal length range, obtain multiple image blocks through an image segmentation method, detect the target in each image block using an image detection model, and then integrate the targets in all image blocks to obtain a detection result.

[0081] It should be noted here that each module in this embodiment corresponds to each step in Example 1 one by one, and the specific implementation process is the same, which will not be repeated here.

[0082] Embodiment 3

[0083] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the long-distance small target detection method described in the above embodiment 1 are implemented.

[0084] Embodiment 4

[0085] This embodiment provides a computer device, such as Figure 2 As shown, it includes a computer-readable storage medium 1003, a processor 1001, a communication interface 1002, and a computer program stored on the computer-readable storage medium 1003 and executable on the processor 1001, wherein the processor 1001, the communication interface 1002, and the computer-readable storage medium 1003 may be connected via a bus or other means. The communication interface 1002 is used to receive and send data, and when the processor 1001 executes the program, the steps in the long-distance small target detection method described in the first embodiment are implemented.

[0086] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0087] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0088] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0089] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for detecting small targets at long distances, It is characterized in that include: Obtain the sensor diagonal size, aspect ratio and resolution, calculate the physical width of the sensor based on the sensor diagonal size and aspect ratio, calculate the physical size of each pixel based on the sensor's physical width and resolution, and convert the pixel width of the target in the image into the physical width of the target; Obtaining the actual width of the target and the distance range between the target and the sensor, and determining the focal length range in combination with the physical width of the target; An image within the focal length range is acquired, and a plurality of image blocks are obtained through an image segmentation method. After the target in each image block is detected using an image detection model, the targets in all the image blocks are integrated to obtain a detection result.

2. A method for detecting a small target at a long distance as claimed in claim 1, It is characterized in that The physical width of the sensor is: IN s =k×Q1 Among them, the scaling factor The resolution is P1×P2, the aspect ratio is Q1:Q2, and the diagonal size is H.

3. A method for detecting a long-distance small target as claimed in claim 1, It is characterized in that The physical width of the target: Where, the physical width of the sensor is W s , resolution is P1×P2, W 1 is the pixel width of the target in the image.

4. A method for detecting a long-distance small target as claimed in claim 1, It is characterized in that The relationship between the focal length range and the distance range between the target and the sensor is: Where f is the focal length, D is the distance between the lens and the target, and W is 3 is the actual width of the target, the physical width of the target is W 2 .

5. A method for detecting a long-distance small target as claimed in claim 1, It is characterized in that The image segmentation method is: segmenting the image into n1 parts in the long axis direction, segmenting the image into n2 parts in the short axis direction, and finally segmenting the image into n1+n2 image blocks.

6. A method for detecting a long-distance small target as claimed in claim 1, It is characterized in that The image detection model adopts YOLOv8, and the backbone network of YOLOv8 adopts MobileNetV3.

7. A method for detecting a long-distance small target as claimed in claim 1, It is characterized in that The data set used for training the image detection model is obtained by annotating images collected at various angles and heights after brightness enhancement, rotation and flipping.

8. A long-distance small target detection system, It is characterized in that include: The target physical width calculation module is configured to: obtain the diagonal size, aspect ratio and resolution of the sensor, calculate the physical width of the sensor based on the diagonal size and aspect ratio of the sensor, calculate the physical size of each pixel based on the physical width and resolution of the sensor, and convert the pixel width occupied by the target in the image into the physical width of the target; A focal range determination module is configured to: obtain an actual width of the target and a distance range between the target and the sensor, and determine a focal range in combination with the physical width of the target; The target detection module is configured to: acquire an image within the focal length range, obtain multiple image blocks through an image segmentation method, detect the target in each image block using an image detection model, and then integrate the targets in all image blocks to obtain a detection result.

9. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the program is executed by a processor, the steps in a long-distance small target detection method as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, It is characterized in that When the processor executes the program, the steps in the long-distance small target detection method as described in any one of claims 1-7 are implemented.

Citation Information

Patent Citations

  • Specific target automatic capture method and device

    CN106327530A

  • ARM-based remote target clear imaging system and method

    CN113792708A

  • Super-resolution enhancement small target detection method based on image partitioning

    CN115082775A

  • Remote small target rapid detection method fusing depth information

    CN115797399A

  • Full image detection

    US20220027642A1