Method And System Implementation Of Edible Rose Disease Detection Based On Deep Learning

Crop diseases have a direct impact on the normal growth of crops,resulting in low land yields and reduced economic incomes.Timely and accurate identification of crop disease types is a prerequisite for accurate application.This study focuses on the difficulty of deploying disease detection algorithms to cultivate edible rose plants.We have developed a leaf of rose’s disease detection model based on an improved Yolov5 network and established a corresponding monitoring and warning system and web application.The system can accurately and timely detect the types of diseases and send alerts to users,providing a basis for disease prevention and precise medication.The specific research work includes:(1)In order to provide sufficient training data for rose disease detection algorithms,we used a combination of image capture and annotation to create a dataset of 565 images of rose leaves with the disease.Crop-related scholars labeled the leaves and lesions in the data as the standard to establish the VOC format raw dataset.Then,data augmentation methods such as Posterize,Sharpness,and Auto Contras were applied to expand the dataset,After expansion,the dataset is divided into 2712 training sets,658 validation sets,and the rest for testing experiments for preservation,resulting in a suitable dataset for training rose disease detection algorithms.(2)In order to address the high cost of manual identification and the lack of disease recognition models for rose leaf diseases,a leaf of rose disease detection model based on Yolov5 s was proposed.The model’s training set and validation set were divided from the rose leaf disease dataset,and improvements were made to the prior box clustering method,bounding box loss function,upsampling method,and feature fusion module based on the original model.The experimental results show that the average precision(m AP)of the improved model on the dataset reached 94.86%,significantly higher than the original model’s 91.56%.Through testing,it can be seen that the proposed rose leaf disease identification model outperforms common target detection models such as Faster R-cnn and Retina Net in the rose leaf disease detection task,and can effectively improve the detection accuracy for rose leaf diseases.(3)In response to the difficulty of deploying disease detection algorithms in agricultural production,we researched the rose disease monitoring and warning system.By investigating user expectations and requirements in actual production environments,we analyzed the feasibility and difficulty of implementing disease warnings,planting area monitoring,and database functions.Based on the feasible parts,we determined the following core functions of the system: image data collection,real-time monitoring of planting areas,and disease situation change information push.To achieve these functions,we used the ESP32-cam as the data acquisition module to collect local video data of the planting site and obtain disease detection results through the leaf of rose disease detection model.In the cloud part,based on the disease detection results,we combined the streaming transmission and message push capabilities of SRS and One Signal services to realize the disease warning and planting area monitoring functions and integrate these functions into a web application.After testing,the real-time monitoring module of the system allows users to view the realtime status of the planting area at any time,and the disease warning module can promptly push disease change information to users,enabling them to grasp the changes in crop diseases in the first time even without opening the web application.The proposed rose leaf disease monitoring model and rose disease monitoring and warning system provide a new approach for the rapid detection and precise medication of diseases in edible roses,significantly reducing manual costs.