【摘要】 为解决挖拔式木薯智能收获机械在作业过程需要快速准确地确定茎秆位置的问题，基于YOLO(You only look once)卷积神经网络提出一种检测速率更快且满足准确率的网络设计(CS-YOLO)。首先，采集并扩增木薯茎秆图像数据集，对样本集进行标注与划分；然后，改进YOLOv1网络结构，利用全局平均池化替代全连接层，并适当调整网络深度和宽度，设计了一种新的网络；最后，对网络进行检测性能试验和对比分析。结果表明：新网络模型尺寸较原网络大小减少约一半，平均每张图像的检测耗时约0.015s,检测速度显著提升；当测试阶段IOU(Intersection Over Union)阈值为0.1时，模型准确率达到了99%,提出的检测方法可满足木薯收获机精准作业要求。研究可为实时、准确地检测田间木薯茎秆位置提供了一种新的思路和方法，也为仿生挖拔式木薯收获机提供了技术支撑。更多还原

【Abstract】 In order to solve the problem that the ‘dig-pull’cassava harvester needs to quickly and accurately determine the position of the steams during the operation process, an improved lightweight object detection method based on YOLO(You only look once) is proposed(CS-YOLO). Firstly, the cassava stem image data set was collected and expanded, and the sample set was labeled and divided. Then, the YOLOv1 network structure was improved, the global average pooling was used to replace the fully connected layer, and the network depth and width were appropriately adjusted. Finally, the detection performance of the network was tested and compared. The experimental results show that the size of the new network model is reduced by about half compared with the original network size, the average detection time of each image is about 0.015 s, and the detection speed is significantly improved. When the threshold of IOU(intersection over union) in the test stage is 0.1, the model’s accuracy reaches 99%. This study provides a new idea and method for real-time and accurate detection of cassava stems in the field and provides technical support for bionic cassava harvester.更多还原