【摘要】 草莓成熟度和空间位置的识别是草莓采摘机器人研究的重要环节 ,解决此问题必须首先对采集的草莓图像进行分割。采用三层BP神经网络 ,通过分析选取 3× 3邻域像素的H通道值作为草莓图像的特征 ;选取HSV模型中与亮度无关的通道以排除图像的明暗对分割效果的影响 ;采用单通道以缩短图像处理时间。选取 2 0幅图像作为训练样本 ,以人工借助Photoshop软件分割后的图像作为教师信号 ,采用BP算法对神经网络的权值进行训练。经过 10 0次循环后 (误差为 0 0 0 1) ,获得了有效的网络权值。试验结果表明 ,利用BP神经网络能较好地实现成熟草莓果实与背景的分离 ,经过提取大区域和腐蚀、膨胀等算法的进一步处理后 ,效果更好 ;而且 ,只要改变训练时的教师信号 ,即可实现对草莓果梗、萼片等图像的分割更多还原

【Abstract】 For the strawberry harvesting robot, the degree of maturation and recognition of the position in space are the important steps. First of all, segmentation of strawberry image is required. The three layers BP neural network method for segmenting the true color of strawberry image was studied. The H channel values of 3×3 neighborhood pixels were obtained as features through analysis. In the HSV model, channel H was chosen to get rid of the effect of light because it had no relation with light. Single channel was selected to shorten processing time. In the network, 20 strawberry images were taken as training samples. And results of manual segmentation images were taken as teacher signals. The BP algorithm was used to train the parameter of the network. The effective parameter was achieved after 100 times’ training. The error was 0 001. The result of this experiment showed that expected segmentation result could be achieved by using BP network method. After taking further processing such as picking up bigger areas, erosion and dilation, etc., the result could be better. Moreover, if the teacher signals were changed, the image segmentation of peduncle or sepal could be realized easily.更多还原