Recognition Of Common Weeds In Northern Field Based On PyTorch Framework

At present,a key issue in the field of agricultural research is how to control the weeds in the field.Chemical weeding is still the main control method for domestic weeds in the field.Although this method is timely,efficient and economical,it is suitable for modern agricultural production crops.However,its widespread use has many drawbacks to the long-term development of the ecological environment,to some extent.It does not match the current concepts of green environmental protection,sustainable development,and precision farming.The purpose of precision agriculture is to increase crop yield and quality,reduce production costs,reduce pollution,and improve environmental quality.Therefore,in order to be able to use chemical herbicides reasonably and effectively,and to avoid large-area spraying without different sprayers,the identification of weeds has become particularly important.Through research and analysis of weed identification technology at home and abroad,this paper selects Pytorch framework and combines AlexNet model to identify weeds.Its main contents are as follows:1.Fully researched and analyzed the related technologies of computer vision technology in weed identification,and deeply understood the recognition framework,analyzed its shortcomings,and finally selected the Pytorch framework to identify weeds.2.Select 12 kinds of weeds data sets such as foxtail,crabgrass,cocklebur and goosegrass;3.Use the median filter to denoise the image.Combining the pre-processed images,the OTSU adaptive threshold segmentation method is used to separate weeds and backgrounds by comparing and analyzing the image segmentation correlation algorithms.4.Based on the Pytorch framework,the dataset was trained and tested using the AlexNet model with an accuracy of 95.65%.5.Combine the model with the actual,design and implement a system for identifying common weeds in the northern fields.Finally,summarize the research results of this paper and put forward the content to be optimized and the focus of future work.