摘要
为解决人工除草耗时长、农药除草污染大的问题,需要更准、更快的杂草识别定位算法帮助除草机器人根除农田杂草。课题小组提出了一种基于Faster R-CNN的农田杂草反向识别改进算法,新算法利用循环一致性生成对抗网络(Cycle-GAN)的图像生成能力以解决训练样本稀缺的问题,同时将Cycle-GAN与快速区域卷积神经网络(Faster R-CNN)混合使用,从而提高杂草识别能力。试验结果表明,该方法在正常拍摄的测试集图片中识别率可以达到95.06%,识别结果优于传统Faster R-CNN的87.59%。该算法具有识别速度快、实时性好的优点,在果园、园林除草等方面具有应用价值。
In order to solve the problems of time-consuming manual weeding and heavy pollution from chemical weed control,more accurate and faster weed recognition and positioning algorithms are needed to help to weed robots eradicate farmland weed.Our study group proposed an improved algorithm for farmland weed reverse recognition based on Faster R-CNN.The new algorithm uses the image generation capability of Cycle General Adversarial Network(Cycle-GAN)to solve the problem of scarcity of training samples.Meanwhile,the combination of Cycle-GAN and Faster Region Based Convolutional Neural Network(Faster R-CNN)improves the ability of weed recognition.The test results show that the recognition rate of this algorithm can reach 95.06%in recognizing the test pictures normally shot,and the recognition result is 87.59%better than that with the traditional Faster R-CNN.The algorithm has the advantages of fast recognition speed and good real-time performance,and furthermore,it has application value in orchard and garden weeding.
作者
张瑞森
万兴鸿
陈子颖
高昕
唐亚南
Zhang Ruisen;Wan Xinghong;Chen Ziying;Gao Xin;Tang Ya'nan(School of Automation and Electrical Engineering Chengdu Technological University,Sichuan Chengdu611730)
基金
国家级大学生创新训练项目(202011116001)
成都工业学院2020年青苗计划一般项目(2020QM085)。
关键词
快速区域卷积神经网络
杂草识别
循环一致性生成对抗网络
实时性
faster regional convolutional neural network
weed identification
cycle general adversarial network
the real time
