Mobile Terminal System Researches And Development For Recognizing And Capturing The Soybean Phenotype Traits In The Field

As an important food crop and economic crop,soybean occupies an important position in Chinese agriculture production.With the continuous expansion of the soybean breeding scale,the cost of field soybean plant phenotype collection is increasing.Breeders must use modern information means to collect field soybean plant phenotypes.At present,many mobile systems can be used to collect plant phenotypes in China,but no plant phenotype collection system can be seamlessly connected with the upstream and downstream of crop breeding.Most of the phenotypic acquisition systems are limited to recording phenotypic images and lack real-time phenotypic feature recognition function.At the same time as image collection,professionals are required to rely on manual measurement and visual recognition to record phenotypic traits,which increases the uncertainty of phenotypic information.According to the requirements of the soybean breeding team,a mobile terminal system for field soybean phenotype recognition and collection was designed and developed in this paper and integrated into the big data platform of computational breeding.The system was tested and optimized for two years.The system is convenient and easy to use,and can seamlessly connect the research work of breeding upstream and downstream,helping the breeding team to achieve the goal of the whole process of information management from experimental design to phenotypic data analysis.The main work of this paper includes:(1)Study the application scenarios of software in the breeding process of soybean breeders,in-depth analysis of the role of software users,clear the requirements of breeding experiment design,field planting,phenotype collection,and data analysis,and design the logical architecture,network topology architecture and functional architecture of the system.In order to meet the needs of different terminals,the system adopts the way of front and back-end separation for design and development.Among them,the Uniapp framework is used on the front-end mobile end,react framework is used on the PC end,and the Spring Boot framework is used on the back end to develop business logic.Algorithms and deep learning models are placed on the algorithm server built by the Django framework,so as to realize the separation of development business logic and algorithm model.Thus,the algorithm model is deployed on the server with stronger performance and higher computing power,so as to ensure that the operation of the system is more stable and efficient,so as to meet the needs of users.This paper divides the system function into eight functional modules and carries on the detailed design and implementation of each module and database.(2)Design the background plate of soybean leaf image acquisition in the field,propose a geometric correction method of leaf image based on the red circle as the correction point,and study the image segmentation of soybean leaf under natural environment and the measurement method of soybean leaf geometric parameters.Thus,soybean leaf area,perimeter,length,and width can be accurately calculated with the lowest time complexity possible.(3)In order to solve the problem of low efficiency of visual observation,an improved ResNet50 soybean leaf shape classification algorithm was proposed in this paper to realize automatic recognition and trait capture of phenotypes in pictures.Experimental results showed that the accuracy of the improved network was increased by 1.8%compared with the original network.(4)An improved YOLOv5s soybean flower target detection algorithm is proposed.The lightweight MobileNetv3 network is used to replace the backbone network CSPDarkNet53 of YOLOv5s with deep separable convolution,and the CBAM attention mechanism is introduced.Experimental results show that the weight parameters volume of the improved YOLOv5s model decreases by 43.8%and FPS increases by 20%when mAP decreases by 0.34%.The mobile terminal field soybean ph enotype recognition and collection system designed and developed in this paper,after two years of use and improvement,basically meet the needs of soybean breeders for phenotype collection and recognition.The data interface between the mobile terminal system with the other functional modules of the platform has been finished.It effectively alleviates the problems of time-consuming and laborious errors in soybean phenotype collection and phenotype recognitions,and meets the needs of breeding experiment designers for the phenotype pre-estimation of experimental materials.So,it really improves the efficiency of soybean computational breeding.