基于轮廓特征的稻种芽谷检测方法

根据机器视觉检测杂交水稻种子质量的要求，针对单粒、静态稻种图像进行芽谷分析识别。对金优402、汕优10、中优207和加优99品种的稻种采集了黑、白背景， A、B两面共4×400幅图像，每幅图像提取出16个稻种轮廓特征参数，经主成分分析降维后作为网络输入，对网络结构进行优化并充分训练后分别建立了各品种的两层人工神经网络。网络对测试集正常稻种的识别准确率均超过95%，对芽谷的识别准确率在85%至90%之间。

Inspection of germinated rice seed on panicle based on contour features

A digital image analysis algorithm was developed to quickly and accurately inspect the germinated rice seed on panicle based on contour features. The algorithm was applied to a 4×200 images set which includes black background, white background and both side images of rice seed. Four ANNs were established for rice varieties: Jinyou402, Shanyou10, Zhongyou207 and Jiayou. The results show that the algorithm achieved an accuracy of 95% above for normal seeds, 85% to 90% for seeds germinated on panicle. Error analysis provided suggestions for increasing the accuracy further.