Research On Counting Method Of Pinus Elliottii Based On Deep Learning

Pinus elliottii is one of the important economic tree species in China,which has the characteristics of fast growth,strong resistance and high quality wood,and is widely planted in Northeast China,North China,East China and Southwest China.Breeding is an important link in the cultivation of slash pine,which is of great significance to improve the production efficiency and quality of slash pine.The shoot number of individual Pinus elliottii plants is an important index for the selection of excellent characters.The traditional counting method of shoot number of Pinus elliottii is mostly done manually,which is not only time-consuming and laborious,but also has low accuracy and efficiency,and the counting process will also cause harm to the plants themselves.With the rapid development of computer technology and image processing technology,it is possible to realize the automatic counting of Pinus elliottii shoots with the help of computer vision technology.Therefore,in this paper,the problem of shoot counting of Pinus elliottii plants in the forest was deeply studied,a shoot counting model was constructed and an automatic shoot counting system was designed to improve the efficiency and accuracy of shoot counting of Pinus elliottii plants,reduce labor costs and provide reliable technical support for the selection of excellent characters of seed examiners.The main work of this paper is as follows:(1)Construct an image data set of slash pine forest based on drone shooting.In order to solve the problems such as the lack of image data set and poor quality of Pinus elliottii forest,this paper takes pictures of wetland pine forest at a fixed height with the help of drone,and screens out clear forest images;Aiming at the complex background and the similarity between the shoots and the branches of Pinus elliottii,the data set is enhanced by filtering to construct a high-quality data set of Pinus elliottii.(2)Build a model for extracting the crown of Pinus elliottii based on YOLOX.In order to accurately extract individual tree crowns from UAV images containing multiple tree crowns,the target detection network YOLOX is introduced and trained on the high-quality data set constructed in this paper to detect the position information of individual plants in the image and generate multi-scale image data sets containing only individual plants.In order to verify the effectiveness of YOLOX model,it is analyzed and compared with several common target detection models on the same data set,and the accuracy of YOLOX model is over 99%,which is better than YOLOv5,Efficient Net and Faster-RCNN networks.(3)Build an automatic counting model of Pinus elliottii shoot quantity based on CCTrans.Aiming at the problem of automatic counting of the final shoot quantity of slash pine,based on the image of slash pine obtained in the research content(2),a shoot quantity counting model of slash pine was constructed.The model takes the crowd counting network CCTrans as the benchmark network,and introduces perspective conduction and unbalanced transmission to improve the original loss,so as to reduce the calculation amount in the network detection process and improve the accuracy of the model.In order to verify the validity of this experiment,the classical counting networks DM-Count,CSR-net and MCNN are used for comparison on the same data set.The error of the improved network in this study is not higher than 3%,which is superior to other counting networks in both speed and accuracy.(4)Build an automatic detection system of Pinus elliottii shoot quantity based on python.In order to facilitate the application of the research results by breeders,an automatic detection system of Pinus elliottii on the web page was developed by using python language,which integrated the extraction function of images of Pinus elliottii in the forest with the detection function of single Pinus elliottii,so as to realize the accurate detection of the shoot amount of each Pinus elliottii on the images collected by drones.