Aerospace image recognition method for image focusing detection based on Gaussian model

Abstract

The invention belongs to the technical field of image processing and recognition, and particularly relates to an aerospace image recognition method for image focusing detection based on a Gaussian model, which comprises the following steps of: acquiring an aerospace image; the collected images are preprocessed; extracting feature information of a target from the preprocessed image by using a deep learning technology, and performing feature extraction on the image by using a convolutional neural network (CNN); a target is detected according to the extracted feature information, and a target detection algorithm based on deep learning is used; and post-processing the result output by the target detection module. According to the invention, the feature expression in the image is automatically learned by using the deep learning technology, and the target can be accurately identified. And through an optimization algorithm and a parallel computing technology, the accuracy of target detection is improved. An efficient algorithm and a hardware acceleration technology are adopted, a large amount of image data can be rapidly processed, and the data processing time is shortened. And meanwhile, the parallel computing technology is adopted to realize computing acceleration, so that the real-time performance is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing and recognition, and particularly relates to an aerospace image recognition method for image focus detection based on a Gaussian model. Background Art

[0002] In the aerospace field, image recognition technology plays a crucial role in achieving various tasks, such as target detection, tracking, recognition, etc. However, due to factors such as small target sizes, diverse shapes, and complex backgrounds in aerospace images, the tasks of target detection and recognition are extremely challenging. Traditional image recognition methods often have low accuracy and large computational amounts when dealing with aerospace images, making it difficult to meet real-time requirements. Therefore, developing an efficient and accurate aerospace image recognition system is a problem waiting to be solved currently. Summary of the Invention

[0003] Aiming at the technical problem that the above traditional image recognition methods often have low accuracy and large computational amounts when dealing with aerospace images and are difficult to meet real-time requirements, the present invention provides an aerospace image recognition method for image focus detection based on a Gaussian model, which can quickly and accurately identify the targets in aerospace images, has high accuracy and real-time performance, and thus helps aerospace to recognize complex images.

[0004] To solve the above technical problem, the technical solution adopted by the present invention is as follows:

[0005] An aerospace image recognition method for image focus detection based on a Gaussian model, comprising the following steps:

[0006] S1. Obtain an aerospace image, and use a high-resolution camera and a high-precision positioning system to ensure that the obtained image has high clarity and accurate positioning;

[0007] S2. Preprocess the collected image, including denoising, enhancement, and color correction operations to improve the image quality. The preprocessing module also includes perspective transformation and coordinate conversion functions to adapt to cameras at different angles and positions;

[0008] S3. Use deep learning technology to extract the feature information of the target from the preprocessed image. Use a convolutional neural network CNN to extract features from the image, which can automatically learn the feature expressions in the image. The feature extraction module also includes multi-scale feature extraction and context information utilization functions to improve the accuracy of target recognition;

[0009] S4. Detect the target based on the extracted feature information. Use object detection algorithms based on deep learning, including YOLO, SSD, or Faster R-CNN, to perform object detection on the preprocessed image. The object detection module also includes functions such as multi-class object recognition and object confidence scoring to achieve multi-object tracking and recognition;

[0010] S5. Post-process the results output by the object detection module, including object tracking, trajectory analysis, and data fusion operations to provide more comprehensive information. The post-processing module also includes object segmentation and instance segmentation functions to improve the fineness of object recognition.

[0011] The method for obtaining aerospace images in S1 is as follows:

[0012] S1.1. Use a camera with a high-resolution CCD or CMOS sensor to improve the spatial resolution of the image. These sensors can capture more detailed details, thereby improving the image quality;

[0013] S1.2. Use an imaging algorithm with super-resolution reconstruction technology to improve the image resolution through information fusion and constrained prior methods;

[0014] S1.3. Use an automatic exposure method based on two-dimensional entropy to optimize the exposure quality through matrix segmentation and curve fitting, ensuring uniform image brightness and clear details.

[0015] The method for denoising the collected image in S2 is as follows: Add noise through the imnoise function, and then use medfilt2 or fspecial('gaussian') for median or Gaussian filtering.

[0016] The method for enhancing the collected image in S2 is as follows: For contrast enhancement, use the histeq function in MATLAB for histogram equalization; Sharpening is achieved through convolution operations. Calculate the gradient using the Sobel operator and apply it through the imfilter function.

[0017] The method for color correction of the collected image in S2 is as follows: Adjust the image brightness to compensate for the non-linear response of the hardware device; Correct the color deviation caused by changes in lighting conditions; Convert the RGB image to a grayscale image or HSV / HSV space for subsequent processing.

[0018] The method for performing perspective transformation and coordinate conversion on the captured image in S2 is: when implementing perspective transformation, it is necessary to determine the coordinates of the four corner points and use the perspectiveTransform function to complete the mapping; the coordinate conversion can be achieved through simple matrix operations, such as using the imrotate or imtranslate function to adjust the image direction and position.

[0019] The method for extracting features from an image using a convolutional neural network (CNN) in S3 is as follows:

[0020] S3.1. Use multiple convolutional layers and pooling layers to extract features of the image layer by layer, from low-level features to high-level features. The low-level features include edges and textures, and the high-level features include object shapes and structures.

[0021] S3.2, introduce residual network ResNet, VGG or GoogLeNet to improve the deep feature extraction capability;

[0022] S3.3, extract features at different resolutions or scales, and achieve fusion of multi-scale information through upsampling or downsampling modules;

[0023] S3.4. Combine the Transformer model or the spatial pyramid pooling SPP module to enhance the model's ability to understand the global context; use multi-scale convolution to extract features at different levels to better capture details in the image;

[0024] S3.5. The extracted multi-scale features are integrated through the fusion layer to form a unified feature representation; finally, the target recognition task is completed through the fully connected layer and classifier.

[0025] The method for implementing multi-target tracking and identification in S4 is:

[0026] S4.1. Use a two-stage detection algorithm to detect targets: Faster R-CNN first generates candidate regions, then classifies and regresses these regions; RPN shares convolutional features with the main detection network to reduce the amount of computation; Faster R-CNN supports feature maps of different scales to improve the detection capability of small targets; the classifier outputs the category probability of each candidate region, and combines the bounding box regression result to generate the final detection result;

[0027] S4.2, Multi-target tracking: Use association mechanisms such as Kalman filter or Hungarian algorithm to match the target detection results in consecutive frames to achieve target tracking; update the state of the target based on its position, speed and other information to ensure the continuity and accuracy of tracking; when the target is occluded or partially lost, restore tracking by re-identifying the target features;

[0028] S4.3. Optimize the model performance by adjusting hyperparameters and introducing regularization techniques.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0030] The present invention uses deep learning technology to automatically learn the feature expressions in images and can accurately identify the targets. By optimizing the algorithm and parallel computing technology, the accuracy of target detection is improved. The present invention adopts an efficient algorithm and hardware acceleration technology, can quickly process a large amount of image data, shortens the data processing time. At the same time, parallel computing technology is used to achieve computing acceleration and improve the real-time performance. The entire system of the present invention realizes automatic control and remote monitoring, reduces manual intervention, and improves work efficiency. By integrating with other aerospace systems, more comprehensive avionics system automation can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary, and for those of ordinary skill in the art, without creative efforts, other implementation drawings can also be obtained according to the provided drawings.

[0032] The structures, ratios, sizes, etc. shown in this specification are only used to cooperate with the content disclosed in the specification for those who are familiar with this technology to understand and read, and are not used to limit the limiting conditions for the implementation of the present invention. Therefore, they do not have technical essential meanings. Any modification of the structure, change of the proportional relationship or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope covered by the technical content disclosed in the present invention.

[0033] Figure 1 It is the step flow chart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0034] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. These descriptions are only for further explaining the features and advantages of the present invention, rather than limiting the claims of the present invention; based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope protected by the present application.

[0035] The following will further describe in detail the specific implementation manners of the present invention in conjunction with the accompanying drawings and embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0036] An aerospace image recognition method for image focus detection based on a Gaussian model, comprising the following steps:

[0037] Step 1: Obtain aerospace images, using a high-resolution camera and a high-precision positioning system to ensure that the obtained images have high clarity and accurate positioning.

[0038] Step 1.1: Use a camera with a high-resolution CCD or CMOS sensor to improve the spatial resolution of the image. These sensors can capture more detailed details, thereby improving the image quality.

[0039] Step 1.2: Use an imaging algorithm with super-resolution reconstruction technology to improve the image resolution through information fusion and constrained prior methods.

[0040] Step 1.3: Use an automatic exposure method based on two-dimensional entropy to optimize the exposure quality through matrix segmentation and curve fitting, ensuring that the image brightness is uniform and the details are clear.

[0041] Step 2: Preprocess the collected images, including denoising, enhancement, and color correction operations to improve the image quality. The preprocessing module also includes perspective transformation and coordinate conversion functions to adapt to cameras at different angles and positions.

[0042] The method for denoising the collected images in Step 2 is: adding noise through the imnoise function, and then using medfilt2 or fspecial('gaussian') for median or Gaussian filtering.

[0043] The method for enhancing the collected images in Step 2 is: for contrast enhancement, use the histeq function in MATLAB for histogram equalization. Sharpening is achieved through convolution operations, calculating the gradient using the Sobel operator, and applying it through the imfilter function.

[0044] The method for color correction of the collected images in Step 2 is: adjusting the image brightness to compensate for the non-linear response of the hardware device. Correcting the color deviation caused by changes in lighting conditions. Converting the RGB image to a grayscale image or HSV / HSV space for subsequent processing.

[0045] The method for performing perspective transformation and coordinate conversion on the collected images in Step 2 is as follows: When implementing perspective transformation, the coordinates of four corner points need to be determined, and the perspectiveTransform function is used to complete the mapping. Coordinate conversion can be achieved through simple matrix operations, such as using the imrotate or imtranslate function to adjust the image orientation and position.

[0046] Step 3: Utilize deep learning technology to extract the feature information of the target from the preprocessed images. The convolutional neural network CNN is used to extract features from the images, which can automatically learn the feature expressions in the images. The feature extraction module also includes functions for multi-scale feature extraction and context information utilization to improve the accuracy of target recognition.

[0047] Step 3.1: Use multiple convolutional layers and pooling layers to perform layer-by-layer feature extraction on the images, from low-level features to high-level features. Low-level features include edges and textures, while high-level features include object shapes and structures.

[0048] Step 3.2: Introduce residual networks ResNet, VGG, or GoogLeNet to enhance the deep feature extraction ability.

[0049] Step 3.3: Extract features at different resolutions or scales, and achieve the fusion of multi-scale information through upsampling or downsampling modules.

[0050] Step 3.4: Combine the Transformer model or the spatial pyramid pooling SPP module to enhance the model's ability to understand global context. Use multi-scale convolutions to extract features at different levels to better capture the details in the images.

[0051] Step 3.5: Integrate the extracted multi-scale features through a fusion layer to form a unified feature representation. Finally, complete the target recognition task through fully connected layers and classifiers.

[0052] Step 4: Detect the target based on the extracted feature information. Use deep learning-based object detection algorithms, including YOLO, SSD, or Faster R-CNN, to perform object detection on the preprocessed images. The object detection module also includes functions such as multi-class object recognition and object confidence scoring to achieve multi-object tracking and recognition.

[0053] Step 4.1: Detect the target using a two-stage detection algorithm. Faster R-CNN first generates candidate regions and then classifies these regions and regresses the bounding boxes. The RPN shares convolutional features with the main detection network, thus reducing the computational cost. Faster R-CNN supports feature maps of different scales to improve the detection ability for small targets. The class probability of each candidate region is output by the classifier, and the final detection result is generated by combining the bounding box regression result.

[0054] Step 4.2: Multi-object tracking. Use association mechanisms such as the Kalman filter or the Hungarian algorithm to match the target detection results in consecutive frames to achieve target tracking. Update its state according to information such as the position and speed of the target to ensure the continuity and accuracy of tracking. When the target is occluded or partially lost, the tracking is restored by re-identifying the target features.

[0055] Step 4.3: Optimize the model performance by adjusting hyperparameters and introducing regularization techniques.

[0056] Step 5: Post-process the results output by the target detection module, including target tracking, trajectory analysis, and data fusion operations to provide more comprehensive information. The post-processing module also includes target segmentation and instance segmentation functions to improve the fineness of target recognition.

[0057] The above only elaborates on the preferred embodiments of the present invention in detail. However, the present invention is not limited to the above embodiments. Within the knowledge scope of those of ordinary skill in the art, various changes can be made without departing from the gist of the present invention, and all such changes should be included within the protection scope of the present invention.

Claims

1. An aerospace image recognition method for image focus detection based on a Gaussian model, characterized in that, It includes the following steps: S1. Obtain aerospace images, using a high-resolution camera and a high-precision positioning system to ensure high clarity and accurate positioning of the acquired images; S2. Preprocess the acquired images, including denoising, enhancement, and color correction operations to improve image quality. The preprocessing module also includes perspective transformation and coordinate conversion functions to adapt to cameras at different angles and positions; S3. Use deep learning techniques to extract the feature information of the target from the preprocessed images. Employ a convolutional neural network (CNN) to extract features from the images, which can automatically learn the feature expressions in the images. The feature extraction module also includes multi-scale feature extraction and context information utilization functions to improve the accuracy of target recognition; S4. Detect the target based on the extracted feature information. Use deep learning-based target detection algorithms, including YOLO, SSD, or Faster R-CNN, to detect targets in the preprocessed images. The target detection module also includes functions such as multi-class target recognition and target confidence scoring to achieve multi-target tracking and recognition; S5. Post-process the results output by the target detection module, including target tracking, trajectory analysis, and data fusion operations to provide more comprehensive information. The post-processing module also includes target segmentation and instance segmentation functions to improve the fineness of target recognition.

2. The aerospace image recognition method for image focus detection based on the Gaussian model according to claim 1, characterized in that, The method for obtaining aerospace images in S1 is as follows: S1.

1. Use a camera with a high-resolution CCD or CMOS sensor to improve the spatial resolution of the images. These sensors can capture more detailed details, thereby enhancing image quality; S1.

2. Adopt an imaging algorithm with super-resolution reconstruction technology to enhance the image resolution through information fusion and constrained prior methods; S1.

3. Use an automatic exposure method based on two-dimensional entropy to optimize the exposure quality through matrix segmentation and curve fitting, ensuring uniform image brightness and clear details.

3. The aerospace image recognition method for image focus detection based on the Gaussian model according to claim 1, wherein, The method for denoising the acquired images in S2 is: Add noise through the imnoise function, and then use medfilt2 or fspecial('gaussian') for median or Gaussian filtering.

4. The aerospace image recognition method for image focus detection based on the Gaussian model according to claim 1, characterized in that, The method for enhancing the acquired images in S2 is: For contrast enhancement, use the histeq function in MATLAB for histogram equalization; Sharpening is achieved through convolution operations. Calculate the gradient using the Sobel operator and apply it through the imfilter function.

5. The aerospace image recognition method for image focus detection based on the Gaussian model according to claim 1, wherein The method for color correction of the acquired images in S2 is: Adjust the image brightness to compensate for the non-linear response of the hardware device; Correct the color deviation caused by changes in lighting conditions; Convert the RGB image to a grayscale image or HSV / HSV space for subsequent processing.

6. The aerospace image recognition method for image focus detection based on the Gaussian model according to claim 1, wherein, The method for perspective transformation and coordinate conversion of the acquired images in S2 is: When implementing perspective transformation, the coordinates of four corner points need to be determined, and the perspectiveTransform function is used to complete the mapping; Coordinate transformation can be achieved through simple matrix operations, such as using the imrotate or imtranslate functions to adjust the orientation and position of the image.

7. The aerospace image recognition method for image focus detection based on the Gaussian model according to claim 1, characterized in that, The method of using a convolutional neural network CNN to extract features from the image in S3 is as follows: S3.

1. Use multiple convolutional layers and pooling layers to extract features from the image layer by layer, from low-level features to high-level features. The low-level features include edges and textures, and the high-level features include object shapes and structures. S3.

2. Introduce residual networks ResNet, VGG, or GoogLeNet to enhance the deep feature extraction ability. S3.

3. Extract features at different resolutions or scales, and achieve the fusion of multi-scale information through upsampling or downsampling modules. S3.

4. Combine the Transformer model or the spatial pyramid pooling SPP module to enhance the model's ability to understand global context; use multi-scale convolutions to extract features at different levels to better capture details in the image. S3.

5. Integrate the extracted multi-scale features through a fusion layer to form a unified feature representation; finally, complete the object recognition task through a fully connected layer and a classifier.

8. The aerospace image recognition method for image focus detection based on the Gaussian model according to claim 1, wherein, The method of achieving multi-object tracking and recognition in S4 is as follows: S4.

1. Use a two-stage detection algorithm to detect the target: Faster R-CNN first generates candidate regions, and then classifies and regresses the bounding boxes of these regions; RPN shares convolutional features with the main detection network, thus reducing the computational amount; Faster R-CNN supports feature maps of different scales to improve the detection ability for small targets; output the class probability of each candidate region through a classifier, and generate the final detection result by combining the bounding box regression result. S4.

2. Multi-object tracking: Use association mechanisms such as Kalman filters or the Hungarian algorithm to match the target detection results in consecutive frames to achieve target tracking; update its state according to information such as the position and speed of the target to ensure the continuity and accuracy of tracking; when the target is occluded or partially lost, restore the tracking by re-identifying the target features. S4.

3. Optimize the model performance by adjusting hyperparameters and introducing regularization techniques.

Images