Research Of A Method Of Intelligently Recognizing Dock Targets

With the development of the remote sensing technology and image processing technology, the recognition of objects of interest in remote sensing images through multi-sources has become an important research topic. As an important military and civilian facility, the dock target has an important strategic significance. It is important to fast and accurately detect dock targets from remote sensing images. It offers the ability to military precisely strike and navigation service. This paper studies the method of intelligent recognition of dock targets. This includes an AIAC (Alterable Included Angle Chain) algorithm for dock detection and target features extraction from multi-source images to achieve the intelligent recognition of dock target.Potential region segmentation method based on AIAC:feature extraction is the key to identify objects, this paper proposed an improved AIAC algorithm. Comparing to the traditional algorithm, this newly developed algorithm can achieve a better performance by retaining the corners with larger curvature value, and removing the slight fluctuation to make the curve smoother. It offers the possibility to use straight lines to simulate curves, and finally replacing curves with polygonal line. And it has positive effectiveness on angle feature and linear feature extraction. This paper used this method to extract the edge features of coastline such as the dock angle, straight line and parallel features. It then, marked the potential area of the dock according to the location relations between dock targets.Dock target recognition method based on the complementary features of multi-source image:Multi-source images can reduce the issues of insufficient information contained in a single image. This paper proposed a method of extracting different features from multi-source images to extract the region of interest, the dock targets, from remote sensing images based on the multi-source features complementation. Firstly, the potential areas were segmented from multi-image. Then the features from multi-source images in the same area were extracted for example the histogram feature and color moment feature in multi-spectral image, and texture feature in full-color image and SAR image. Finally these features were used to train the SVM classifier to classify the true dock target areas from all potential areas. In this paper, a new dock target segmentation and recognition method has been designed to realize the target recognition. The experimental results have shown that the recognition accuracy of convex dock is 86%.