为解决酿酒葡萄生长状态的在线自动监测问题,该文提出了一种基于卷积神经网络的葡萄叶片检测算法.通过多层卷积的方式产生特征图,使原图像的特征增强并且降低了图像噪声,在最后一层特征图中,通过使用RPN(Region proposal network)生成建议区域,然后进行池化操作,最后进行边框回归与分类.该算法在有叶片遮挡、光照阴影、病害叶片等复杂背景因素下对葡萄叶片有良好的检测效果.试验表明,该算法在复杂背景下对葡萄叶片的检测率为87.2％,误检率为7.2％.%It would be necessary to detect grape leaves automatically for the purpose of solving the monitoring issue that focused on the growing status of wine grape.This paper aimed at the detecting method for grape leaf based on deep neural networks algorithm.Such algorithm enhances the original image feature and reduces the image noise through the way of multideck convolutional generates the feature map.On the last feature map,region proposal network could generate proposal,make pooling and bounding box regression and classification as well.Furthermore,such algorithm could produce a relatively obvious effect especially under the complicated circumstance that included the grape leaf shielding,lighting shadow and leaf disease.After the test,this algorithm achieved 87.2％ detection rate and 7.2％ error rate for grape leaf under rigorous situation.
展开▼
