水果自动识别的BP神经网络方法.pdfVIP

图形、图像与多媒体 Image Processing and Multimedia Technology 水果自动识别的BP 神经网络方法* 陈 源 张长江 , （浙江师范大学 数理与信息工程学院， 浙江 金华 321004 ） 摘 要 针对多种水果混合的图像， 对各种水果的提取和识别进行研究。 利用 软件进行图 : Matlab 像数据获取、对比度增强、去噪、二值化处理；为弥补二值化后图像中出现的断边、孔洞，借助 Sobel 算 子进行边缘提取以接合断边， 并基于数学形态学算子填充孔洞。 再标签化处理图像并提取水果的颜 色、形状、边缘特征，用200 幅水果图像提取上述特征构造训练样本和测试样本，然后利用构造的样本 对 神经网络进行训练和测试。 实验结果表明 所提出的方法能够获得很高的正确识别率，能够有效 BP , 地将同一幅图像中的不同水果识别出来。 关键词: BP 神经网络；水果识别；二值化；标签化；MATLAB 中图分类号： TP389.1 文献标识码： A 文章编号： 1674-7720(2010)22-0040-04 Automatic recognition for fruit based on BP neural network CHEN Yuan, ZHANG Chang Jiang (College of Mathematics, Physics and Information Engineering, Zhejiang Normal University, Jinhua 321004, China) Abstract ： The effective recognition of the fruit is the key of fruit automated sorting. The algorithm to extract and identify dif- ferent fruits in the same image is studied. Firstly, some pre-processing which include data acquisition, contrast enhancement, noise removal and binary for the fruit image are implemented by MATLAB software. In order to make up the broken edges and holes in the binary image, Sobel edge detection is used to joint the broken edges, and mathematical morphology operator is employed to fill holes. Labeling is implemented to the pre -processed image in order to extract the fruits color, shape, edge characteristics. The training and testing samples are constructed by extracting the fea