Research On Soybean Emergence Detection Algorithm Based On UAV Aerial Images

Soybean,the primary main source of protein in food,is one of the main cereal food crops in China.In recent years,the demand and import volume of soybeans in China have increasing year by year,and the work of cultivating high-quality and high-yielding soybeans has become an urgent task.The emergence traits of soybean are one of the key contents of phenotypic genomics data in soybean breeding experiments.In soybean breeding,the traditional statistics of emergence rate is that breeders conduct on-the-spot investigation and estimation at the emergence stage to obtain approximate statistical data.With the increase in experimental planting area,the cost of this work has increased,and the data with uncontrollable accuracy cannot support soybean computational breeding.Aiming at the problem that it is difficult to accurately count soybean plants in the field in the breeding experiment,according to the needs of the breeding team,this paper attempts to use the object detection technology to process the aerial images of the drone to realize the accurate statistics of phenotypic traits such as the soybean emergence rate and the number of reserved seedlings per unit area.This paper uses the aerial images of soybean experimental fields planted by the Northeast Institute of Geography and Agroecology,Chinese Academy of Sciences as the basic data set for research,proposing a soybean seedling plant detection model based on YOLOv5.The model can not only be used for statistics of seedling emergence in breeding experiments,but also provide data support for the activities of replenishing seedlings for lack of seedlings in precision agriculture.Based on this method,a long-term monitoring mechanism for experimental fields can be established to provide a reference for the collection of more crop phenotypic traits from UAV aerial images.The main work content of this paper includes:(1)UAV aerial photography technology was used to obtain growth images of soybeans at different periods of emergence.A total of 120 high-resolution images were collected at a height of 5 meters from the ground,and a total of 2,400 source images of the dataset were obtained by segmenting 15%overlapping pixels.Complete the production of the dataset through the current mainstream image processing process;(2)Aiming at the problem of small high-score images,select the current mainstream single-stage target detection representative model YOLOv5,add SPPF module,CBAM attention mechanism module to the network model,optimize EIoU and Focal loss function to improve detection accuracy,and use reasonable pre-training strategy While solving the problem of insufficient data sets,the model optimization reached 97.9%mAP,which was 3.1 percentage points higher than the original model;(3)Aiming at the insufficient amount of UAV aerial images caused by the planting area of the test field,pixel segmentation is used to overlap and crop high-resolution large images,and complete the task of expanding the data set while removing some complex backgrounds.Based on UAV aerial images,this paper uses object detection technology to realize the accurate collection of soybean emergence rate and other traits,the collected data is integrated into the breeding big data platform,completing an important link in the information management of the whole process of soybean computational breeding,effectively improving the accuracy and efficiency of field phenotype collection.