基于卷积神经网络的蔗芽识别定位

Recognition and Location of Sugarcane Bud Based on Convolutional Neural Network

蔗芽识别定位是甘蔗防伤芽切种和甘蔗智能化定向种植的前提。为此,以识别蔗种图像中的蔗芽及定位蔗芽形心位置为目标,构建蔗种图像库;基于蔗芽平面位置识别要求,通过对经典LeNet-5网络结构进行改进得到适合蔗芽平面位置识别的卷积神经网络模型,包括1层卷积层、1层归一化层、1层池化层和1层全连接层。卷积神经网络结构基于MatLab实现,对150幅图像进行实验结果表明:网络模型对蔗芽位置的识别准确率达92%,训练时间为0.8h,模型规模为68.35MB,单幅蔗种图像检测平均耗时1.2s。该方法可有效降低蔗芽的误识别率,防止蔗芽误识别引起的伤芽现象和种植过程中蔗芽朝下引起的发芽率低的现象,可为甘蔗的智能化精准种植提供研究基础。

Directional planting of sugarcane is the future development trend.Accurate identification and positioning of sugarcane buds is the prerequisite for sugarcane anti-damage bud cutting and intelligent directional planting of sugarcane.To identify the centroid position of the sugarcane buds in the sugarcane seed image,construct the sugarcane seed image library;based on the task requirements for sugarcane bud plane position recognition,a convolutional neural network model suitable for sugarcane bud plane position recognition is obtained by simplifying the LeNet-5 network,including a convolutional layer,a normalization layer,and a pooling layer and a fully connected layer.The structure of the convolutional neural network is implemented based on MATLAB.The experimental results show that the accuracy of the network for the position recognition of the sugarcane buds in the sugarcane seed image is 92%,the training time is 0.58h,the model scale is 1.35MB,and the single cane seed image is detected the average time is 1.2s.The method can effectively reduce the false recognition rate of sugarcane buds,prevent the damage to the buds caused by the false recognition of the sugarcane buds and the low germination rate caused by the sugarcane buds facing downward during the planting process,and provide a research foundation for the intelligent and precise planting of sugarcane.