基于神经网络和图像颜色、形状特征的绿色苹果图像分割

Research on the Segmentation of Apple Recognition Based on BP Neural Network and Image Color,Shape Features

图像分割是苹果采摘机器准确识别和定位苹果的关键步骤。本研究首先采用线剖面方法对采集的苹果图像针对颜色特征进行分析,提出了利用颜色特征R-B的色差法对青果期苹果图像进行初步分割。在利用分割后的图像提取图像区域的形状特征(面积、周长、圆形度、离心率等)。然后将得到的8个形状特征作为BP神经网络的输入量,随机选取一定数量的样本图像作为BP神经网络的训练样本图像和验证样本图像。样本图像经过BP神经网络训练后,建立了绿色苹果图像的分割模型。通过BP神经网络分割后的苹果图像,果实识别率高达89.3%,分割效果良好。

Image segmentation is the key step of identification and location accurately for apple picking machin.In this paper the R-B color features work on the fruit of apple for preliminary segmentation was proposed by using the line profile method for collecting apple image to analyse color characteristics.The shape features of image region(area,perimeter,circularity,eccentricity,etc.) were extracted by using preliminary segmentation image.Then,the resulting eight shape features are used as the input to BP neural network,a certain number of sample images were randomly selected as BP neural network training and validation sample image,and a green apple image segmentation model was established after training by BP neural network.The apple recognition rate is 89.3% after using BP neural network segment apple fruit image.