Research And Design Of Adaptive Image Segmentation And Parallel Mining Algorithm

Image segmentation is the first step of image analysis and pattern recognition. The segmentation of interested objects in images has been one of the difficult research problems in image processing. The key of detection and recognition of interested objects in images is the extraction of image features. Remote sensing image is formed by the solar radiation reflected by environment features and the electromagnetic wave signals emitted by the objects itself. With the development of science and technology, remote sensing image plays an important role in many fields, such as industry, production, aviation and so on. Water bridges and airport are important targets in military field and civil field. The segmentation and recognition of water bridges and airport are of great significance in all aspects of society. In this paper, we conducted the research to segment and recognize the airport and water bridges, this include divided into adaptive segmentation, potential region extraction, parallel feature extraction and target recognition.Firstly, an adaptive fuzzy threshold image segmentation method, based on the gray intensity histogram, has been proposed. It obtains the threshold value through the calculation of intensity histogram subsection and histogram inverse transformation. This method was proposed to solve the problem aroused in traditional fuzzy threshold segmentation method, the window width cannot be automatically obtained, and the difficulty in to segmenting objects from images that have single peak or unobvious double peak histogram. This method realizes the adaptive selection of window width, effectively reduces the problems of the traditional fuzzy threshold segmentation method and expands the application field of traditional fuzzy threshold segmentation method. Comparing to the existing related methods, the experimental results have shown that the proposed method can segment objects from the remote sensing images of multi-spectral image, full-color image and SAR image that have single peak or unobvious double peak histogram in a high accurate and efficient way.The key to recognize specific objects from remote sensing images is to extract valid features from these images,In this paper, a parallel feature mining method based on the OpenMP model is proposed. It extracted features, such as geometric invariant moments, texture, color, edge, line, scattering characteristics and geographic information distribution at first. Then, the model was trained by using optimized conjugate gradient method through the BP neural network. The relationships between extracted features were mined according to the results corresponded values of accuracy. Finally, the airports and bridges in remote sensing image can be recognized. According to the experimental results, it is has showed that the parallel mining of extracted features not only reduced the target recognition time, but also effectively improved the recognition accuracy.