苹果识别是开发苹果采摘机器人的关键环节, 利用图像处理技术和神经网络分类器探索苹果图像分割算法。从苹果树图片中选取苹果图像样本和背景图像样本, 分别计算这两类图像样本的颜色特征和纹理特征。颜色特征的计算基于RGB色彩模型, 纹理特征的计算基于灰度共生矩阵。选取适当的颜色特征(R/B值)和纹理特征(对比度值和相关性值)作为输入节点, 利用反向传播神经网络分类器建模, 输出值是一个0～1之间的计算值。通过阈值将输出结果分类为苹果或背景。试验结果表明, 该算法正确率大于87.6%, 对光照的影响不敏感, 是一种较为实用的苹果分割算法。

To improve the accuracy of the automatic detection and classification of apples on the tree, image features and artificial neural network classifier are applied to segment the apple images. First, apple image samples and background image samples are chosen. Then the color feature and the texture features of the samples are calculated. The color feature (R/B ratio) is calculated based on RGB color model, and the texture features (contrast and correlation) are calculated by gray level co-occurrence matrix (GLCM). These three parameters are used as the input to the back-propagation neural network (BPNN) classifier. The result of the output layer is a numerical value in the runge of 0~1. It is classified into fruit and background based on a certain threshold value. The results of the segmentation show that the success rate is over 87.6%, and the influence of light is neglectable. It is feasible to use the algorithm in practical recognition of apple.