Research And Implementation Of Target-oriented Image Feature Mining And Feature Extraction Method

With the rapid development of remote sensing technique, how to effectively extract target feature and mine target feature from remote sensing image has become an important subject of the remote sensing image processing. Because of the importance of the roads, the road feature extraction and mining has great significance in every aspect of social development. In this paper, the main research work is as follows.There is inevitable noise interference in the process of image acquisition or transmission. Noise degrades image quality and brings difficulties in image segmentation, analysis and so on. The paper presents a method of image-denoising based on Hilbert-Huang transform, uses the bi-dimensional empirical mode decomposition to decompose the remote sensing image in multi-scale and acquires the intrinsic mode functions, transforms the intrinsic mode functions by bi-dimensional Hilbert transform to retain and enhance image detail information and weaken the noise. The method achieves the image-denoising result, and the denoising effect can be objectively evaluated.During the feature extraction, color, texture and shape features are extracted. In the edge feature extraction, the paper presents an algorithm, combining with Canny operator and wavelet transform edge detection algorithm. The algorithm combines Canny edge extraction operator and wavelet transform. The Canny edge extraction operator on roads will appear discontinuous phenomena and wavelet transform extraction the road edge will appear false edge information, The algorithm can improve image edge detection.Using BP neural network to train image features, when given a trained network to provide new models of learning and memory, it will disorganize the connection weights, result in disappearance of memory in learning mode of information. In the BP neural network optimization, the main method is to add the momentum factors, change its incentive functions. In this paper, the BP neural network can combine with simulated annealing method, it can improve the effect of learning neural network and shorten the time of falling into local minima.