High-Speed Ship Detection in SAR Images Based on a Convolutional Neural Network -Deep Learning Projects-Matlab

6,000.00 Exc Tax

High-Speed Ship Detection in SAR Images Based on a Grid Convolutional Neural Network -Deep Learning Projects-Matlab
Platform : Matlab

Delivery : One Working Day

Support : Online Demo ( 2 Hours)

100 in stock

SKU: High-Speed Ship Detection in SAR Images Based on CNN Category:

Description

High-Speed Ship Detection in SAR Images Based on a Convolutional Neural Network -Deep Learning Projects-Matlab

ABSTRACT

Small target recognition is a classic challenge in the field of target recognition. In this paper, an improved algorithm based on Yolo V3 is proposed for ship detection in remote sensing images. In order to solve the problem that the input satellite image size is too large, the window sliding segmentation technology is used to cut the large image into several small images; because the detection target only has a single kind of ship, K-means algorithm is used to re cluster anchor Box, to improve the detection accuracy of the algorithm; in view of the small ship target in the ocean background, the multi-level feature extraction method is used to highlight the feature vector of small target to improve the recognition ability of small target. Through the above-mentioned methods, using the data set of Airbus ship detection provided by kaggle, the ship target detection has achieved a better comprehensive accuracy rate, which is higher than the accuracy and recall rate of similar target detection algorithms.

INTRODUCTION

There are many hot spots and sensitive areas in China’s waters, and foreign ships frequently violate and create conflicts. If they can detect and warn in advance, they will have more initiative in the disposal. With the increasing resolution of remote sensing satellites and the increasing transmission bandwidth of satellite-ground link, a large number of HD satellite image resources provide a good solution. Image recognition technology can be applied to remote sensing images, and timely detection of ships in hot spots can be carried out by taking advantage of technology and resources. The application of image recognition to ship intelligent detection has the advantages of high efficiency, high reliability and all-weather operation. However, cloud and land background, complex sea conditions and small ship targets also restrict the effectiveness of the detection model. Faced with these problems, traditional detection methods such as SIFT, HOG and SVM can no longer achieve good detection results. With the rapid development of deep learning in the field of target detection in recent years, the target detection algorithm based on neural network shows good performance, image recognition algorithm based on neural network can be divided into two categories: 1. The two stage method, sample first generate candidate box, and then processed by convolution neural network, the method detection accuracy and better positioning accuracy, representative algorithms such as fast R- CNN, faster R-CNN . 2. The method of one stage directly transforms the problem of target border positioning into a regression problem, which is then processed by the convolutional neural network. This method has a relatively fast operation speed and represents algorithms such as SSD, YOLO, YOLO V3. The width range of remote sensing image is 10km-30km, faced with such a huge shooting range, the ship occupies only a few pixels in the image, to realize fast and accurate target recognition, it is necessary to consider the universality of the algorithm and detection speed. So this paper chooses YOLO V3 detection algorithm for the detection of ship targets, and according to the actual situation to make improvements to the algorithm.

PROPOSED METHOD

In this paper, an improved algorithm based on Yolo V3 is proposed for ship detection in remote sensing images. In order to solve the problem that the input satellite image size is too large, the window sliding segmentation technology is used to cut the large image into several small images; because the detection target only has a single kind of ship, K-means algorithm is used to re cluster anchor Box, to improve the detection accuracy of the algorithm; in view of the small ship target in the ocean background, the multi-level feature extraction method is used to highlight the feature vector of small target to improve the recognition ability of small target.

CONCLUSION

Aiming at the difficult problem of ship identification in remote sensing image, this paper proposes the YOLO V3 based identification method. The algorithm was optimized through training sample prescreening, Anchor box recombination class, feature pyramid network optimization, sliding segmentation of input large image window and other methods, so that the neural network could better extract the ship feature vector, realize accurate identification of ship target in remote sensing image, and reduce the situation of missed detection or false detection. For complex situations such as cloud cover, sea and land interference, and ocean background interference, negative sample enhancement learning method is adopted to add cloud, land background, diverse ocean background and other images into the training set as negative sample enhancement algorithm’s robustness. The experimental results show that the proposed method not only realizes the ship identification in remote sensing images, but also effectively reduces the problems of ship missing detection and false detection caused by the un-favorable factors such as cloud cover, and effectively improves the algorithm performance.

Reviews

There are no reviews yet.

Be the first to review “High-Speed Ship Detection in SAR Images Based on a Convolutional Neural Network -Deep Learning Projects-Matlab”

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.