摘要
冰糖橙图像识别是研究冰糖橙采摘机器人的关键技术。研究以自然光源下用数码相机拍摄的冰糖橙图像为研究对象，以图像3×3邻域的H通道分量值作为图像特征，随机选取30幅图像作为BP神经网络的训练样本，10幅图像作为测试样本。采用Photoshop软件对训练样本进行分割后的结果作为期望输出，迭代次数为150次，误差为0．001，经过训练得到了有效的权值。接着采用BP神经网络将冰糖橙果实与其背景进行分割，识别出冰糖橙果实部分。然后，对分割后的图像进行腐蚀、膨胀等形态学处理；最后，进行Hough圆变换检测，得到果实的外包络圆及圆心位置。
The image recognition of candy orange is the key technology to study the picking robot of rock candy orange. In this study, we used the image of ice orange was taken with digital camera under natural light source as the object. Taking the H-channel component of the image 3×3 neighborhood as the image feature, randomly selecting 30 images as training samples of BP neural network, and selecting 10 images as test samples. Using Photoshop software to split the training samples as the desired output, the number of iterations 150 times, the error is 0. 001, after training to get a valid weight. Then the BP neural network was used to segment the orange fruit and its background, and the fruit part of the orange sugar was identified. Then, the fractured image is corroded and expanded. Morphological processing is carried out. Finally, Hough circle transform is used to detect the outer envelope and center of the fruit.
出处
《机械设计与研究》
CSCD
北大核心
2018年第1期36-38,共3页
Machine Design And Research
