Research And Experiment On Soybean Flower And Pod Identification Devic

For agronomic experts on soybean flowering and pod drop phenotypic surveys there exist methods that rely on the gauze net method or manual observation at intervals [1],which have problems such as poor real-time and subjectivity.The design of a field soybean pod image acquisition device,through field experiments,based on the improved yolov5 algorithm to identify soybean pods during the flowering and podding period in the field,for the flowering and podding period morphological changes,mutual shading of leaves,small target buds and young pods,the Bottleneck CSP structure is modified to reduce the number of modules to retain more shallow features,enhance feature expression,and The CA attention mechanism was introduced into the backbone network to capture direction and position-awareness information for more accurate localization and identification of target areas of interest,and the anchor box size was modified to improve the accurate identification of small target buds and young pods,so as to build the best field pod identification model for this species in this region.The results of the study can effectively replace the manual task of identifying soybean pods in the field.This application will facilitate the study of the basic laws of flower and pod drop and provide an important phenotypic investigation technique for breeding high-yielding varieties with low flower and pod drop rates.The main research components and results are as follows.Pod identification device design.According to the agronomic requirements and field device guidelines,the image acquisition mechanism,control device and image acquisition page were designed,and the mechanical device was used as the basis for controlling the shooting device to obtain single soybean pod images to provide a data set for effective pod identification,as well as providing basic data for the pod platform and providing a basis for platform recall and identification at a later stage.Construction of a soybean pod model.A dataset of 17,048 images,containing more than10,700 pod targets,was created independently to address the lack of a dedicated database of soybean pod images in the field.The dataset was mainly collected from the field environment,with the main problems of leaf shading,flower and flower mutual shading,and flower and pod mutual shading,to increase the generalisability of the model training.Build the YOLOv5 algorithm implementation environment,data enhancement and annotation of the collected image dataset,apply the improved YOLOV5 based flower pod recognition model to effectively identify the complex environment in the field,and construct a field flower pod recognition model suitable for the region and variety with confidence and average accuracy as evaluation indexes,and test the recognition results to verify whether the model meets the soybean field recognition The model was tested to verify that it met the functionality of the soybean field identification device.Soybean pod recognition device testing and platform implementation.Based on the front-end camera of the flower pod device,the images of single soybean plants were collected at regular intervals through a computer-based collection page.The data analysis function design and the backend management function design module can be used to provide a basis for breeding experts.