Research On Spherical Target Recognition Technique Based On Optional Vision

In the research of computer vision and pattern recognition, target identification technique is a hot topic. This thesis is a research focusing on the detection and recognition of images of spherical target underwater using computer vision and pattern recognition.The core problem of target recognition is segmenting and extracting feature information, followed by identification and classification of it. In the fields of man-made targets underwater, spherical target is an important kind. Among many methods of segmentation and edge detection, sub-pixel edge detection method based on Zernide Orthogonal Moment (ZOM) shows excellent function. Towards step line and curved edge, this thesis employ improved ZOM algorithm to achieve edge detection of spherical target by detecting position and direction of the edge of step line.After detecting the position and direction of edge points, we use Random Hough transform to classify them into inliers and outliers according to normal number of distance between target edge and candidate edge. Experiments results show that generated edge matches well with actual target model.More attention should be paid to image feature to recognize general spherical target. Invariant moments can be extracted from the target images because they are independent of translation, scaling and rotation factor. Then the architecture and parameters of the neural network are set by improved BP algorithm, also the networking training is conducted simultaneously, experiment results show its effectiveness of spherical target recognition.