Research On Robot Lawn Weed Recognition Algorithm

Weeds have a great impact on the growth of farmland and ornamental lawn,and need timely and effective control to benefit the health and beauty of the lawn.Among the existing technologies,there are still many problems in the practical application of most weed identification technologies,such as the small sample size of the traditional machine learning algorithm,the low accuracy of the visual recognition system,and the deep learning network has not appeared in the weeds application.In recent years,deep learning has made breakthroughs in the field of target detection and recognition,especially in training speed and accuracy.In the research of better target detection system,the main content of most improvement methods is to use deeper network to achieve the speed and accuracy of target detection and classification.Based on the above discussion,this paper optimizes the feature extraction of data for the first time.Aiming at the problems of complicated lawn weed samples and low identification accuracy,a lawn weed identification algorithm based on the Faster R-CNN+GAN network has been proposed.The improved Faster R-CNN was used to improve the robustness of the network system by adding noise interference generated by GAN,so as to realize end-to-end training of lawn weed detection and recognition system in complex background.The main research contents and achievements of this paper mainly include the following aspects: Firstly,the preset structure design and visual design of the weeding robot system has been introduced.Then lawn data sets have been collected and processed.following the format of PASCAL VOC,the XML files about the lawn data have been generated.Finally,aiming at the problems of inaccurate identification and location of weeds and poor real-time performance of the current weeding robot,a grass weed identification algorithm based on Faster R-CNN+GAN has been proposed.In this method,the Faster R-CNN algorithm was firstly used to train the initialization model,and then the characteristics of data were optimized by adding GAN layer after pooling layer,so as to improve the robustness of the network.The experimental results show that the method is better than the traditional machine learning method,and has the characteristics of high recognition speed.The method can be used for real-time detection and have application value in garden weed cleaning.