基于神经网络的无人机农药喷洒区域智能识别

Neural Network-Based Intelligent Recognition of Pesticide-Treated Areas for UAV Sprayers

为了减少使用无人机(UAV)进行农药喷洒时存在的过施问题,对无人机喷洒区域的智能识别系统进行研究。该系统基于卷积神经网络模型,将航拍采集的农田图片作为样本进行训练、验证、测试,从识别率和处理时间两个方面出发对模型参数进行寻优。试验结果表明,所建立的智能识别系统的平均识别率为92.3%,单张图像平均处理时间为31 ms,可以在满足无人机硬件条件的约束下,实现对喷洒区域的智能识别。

In order to reduce the over-application of pesticide,an intelligent recognition of spraying areas for UAV sprayers was modelled on a convolutional neural network,and trained,verified and tested with farmland images collected from aerial photographs.The model parameters were optimized from the aspects of recognition rates and processing time.The test results showed that the average recognition rate of the intelligent recognition system is 92.3%,and the average processing time of a single image is 31ms.The intelligent recognition of spraying areas can be realized under the constraint of the UAV hardware conditions.