基于Faster-RCNN网络的蔬菜幼苗识别检测

Recognition and Detection of Multi-category Vegetable Seedlings Based on Faster-RCNN Network

田间杂草容易对蔬菜生长产生不利影响,快速准确的检测蔬菜幼苗并去除杂草对提高蔬菜产量和质量有较大影响。针对复杂农业环境下常规蔬菜幼苗识别方法存在的识别精度低、检测速度慢等问题,本文将Faster-RCNN模型引入到蔬菜幼苗识别检测中,先采用Resnet50残差网络作为前置基础网络提取作物特征,然后将特征送入候选区域建议网络进行先验框调整,最后通过感兴趣区域池化网络和全连接层完成目标分类定位。将检测完成的蔬菜幼苗检测模型部署在NVIDIA Jetson TX2嵌入式平台进行测试,蔬菜幼苗平均识别率达到93.92%,平均检测时间为34.4 ms,具有识别速度快和准确率高等优点。本方法可以为后续农业智能装备精准作业所涉及的蔬菜幼苗检测问题提供新方案。

Vegetable production is closely related to our daily life,but in the process of vegetable growth,weed damage will have a great adverse impact on crop yield and quality.If weeds can be quickly and accurately detected and removed during the crop growth cycle,vegetable production will be greatly improved.Considering the shortages of the existing methods,this paper carries out Faster-RCNN model on vegetable seedling identification.Resnet 50 residual network is adopted to extract the characteristics of crops.The final feature layer of resnet50 is sent to the RPN network to adjust the a priori box and suggestion box,which complete the target the target classification and positioning.The training model was deployed in NVIDIA Jetson TX2 embedded system for testing,and the average recognition rate of vegetable seedlings reached93.92%,and the average detection time was 34.4 ms,which had the advantages of fast recognition speed and high accuracy.Its performance can provide a new solution for the detection of vegetable seedlings involved in the precise operation of vegetable agricultural intelligent equipment.