基于BP神经网络的图像目标识别算法研究

Research on Image Target Recognition Algorithm Based on BP Neural Network

图像目标识别系统主要由四部分组成:图像预处理、图像分割、图像描述、分类决策。文章采用基于不变矩的提取特征值方法,这样整个模式识别主要过程包括图像采集、图像预处理、运动目标不变矩提取和分类识别。利用阈值分割得到运动目标图像,再经滤波去噪得到预处理图像,然后对得到的预处理图像分别提取其Hu不变矩,最后采用最小距离分类器和BP神经网络分类器进行分类,判别运动目标类别。

Image target recognition system is composed of four parts,i. e. image preprocessing,image segmentation, image description, and classification and decision. As the method of feature extraction based on invariant moment values is adopted,the main process of whole pattern recognition includes image acquisition, image preprocessing, moving target invariant moments extraction and classification recognition. The threshold segmentation is used to obtain the motion object image,the preprocessing image is obtained by filtering denoising,and the preprocessing images are extracted from Hu invariant moments. Finally,the classification is performed,using the minimum distance classifier and BP neural network classifier,to determine the moving object categories.