Research On Early Crop Disease Detection Based On Infrared Thermography And Deep Learning

Crop disease early detection is a crucial component of field management that can have a significant impact on crop yield.Its ability to "detect early and treat early" can reduce crop losses and improve crop quality by identifying diseased plants early and applying localized pesticide spraying to suppress early growth of pathogens,thus reducing subsequent pesticide use.Once crop diseases enter the disease period,they spread rapidly,making prevention and control more difficult and costly,making early detection of diseases particularly important.With the development of artificial intelligence,machine vision and other technologies are increasingly being applied to the agricultural sector,especially in the field of disease detection.Infrared thermal imaging technology has also been extensively studied in the field of agricultural disease identification.This article proposes a tomato disease detection and recognition system based on infrared thermal imaging and deep learning with the goal of achieving early disease identification.It solves the problem of subjective misjudgment and crop structure destruction caused by early crop detection,and also overcomes the disadvantages of artificial misjudgment,reliance on expert experience,and high cost by using deep learning to solve the problem of disease identification.The specific research work and main conclusions are as follows:(1)The principle and characteristics of infrared thermal imaging technology are explained,and the principle and characteristics of using infrared thermal imaging for early detection of plant diseases are introduced.The current research status and existing problems of infrared thermal imaging and deep learning technology in agricultural production are reviewed.The detection principle of infrared thermal imaging technology and the characteristics of the crop disease recognition part are explained.(2)In order to prevent the interference of diseases on the experimental results,in this study,in order to eliminate the influence of diseases and pests,the image was taken by planting tomatoes,and in order to expand the image data and enhance the generalization,the infrared thermal image of tomato leaves in the greenhouse environment was added,and the data augmentation method was used to expand the data set.The study also used the Label Img tool for sample annotation and prepared a VOC format based infrared thermography dataset of tomato leaves.I In addition,by studying the different temperature and leaf changes in the early disease detection of different crops,in order to improve the level of early detection of diseases,the disease spot images were also classified,and by studying the infrared thermal imaging of disease spots of different crops,the disease spot dataset was divided into 0～0.2 °C,0.2～0.4 °C,0.4～0.6 °C,0.6～0.8 °C and 0.8 °C + five gradients to conduct a comparative study,and finally optimize the quality of the dataset through data enhancement to establish the early disease dataset of tomatoes.(3)Based on the tomato disease infrared thermography dataset,the performance of YOLOv3,YOLOv4,SSD and YOLOv5 algorithms in tomato disease detection was compared,and the results showed that the YOLOv5 algorithm had the best detection performance,and the m AP of the improved YOLOv5 algorithm was 92.2%,which was 1.2%,5.3% and 15.5% higher than YOLOv3,YOLOv4 and SSD networks,respectively.Through the comparison of different data augmentation methods by YOLOv5,the conclusion is the same as in Chapter 2,that is,the image quality after Gaussian filtering is the highest,with m AP values of 93.6%,which is 0.7%,1.2% and 4.5% higher than that of median,mean and non-local average filtering.For YOLOv5 networks,the dataset enhanced by Gaussian filtering has higher accuracy and lower loss,and the m AP is 1.4% higher than that of the first data set processing,and has faster convergence.(4)Aiming at the problem of poor classification and recognition accuracy of YOLOv5 algorithm,this paper proposes an improved YOLOv5 algorithm.Improvements include the introduction of SE attention mechanisms and SPD Conv for handling low-resolution and small objects,and the replacement of methods using SPPF.The results show that the m AP value of the improved YOLOv5 algorithm is95.7%,which is 2.1% higher than that of the pre-improvement network,and the network is also improved compared with the improved network for tomato disease detection under different temperature gradients,and the five gradient m APs are 91.0%,91.6%,90.4%,92.6% and 94.0%,respectively,which are higher than the 3.6%,1.5%,7.2%,0.6% and 0.9% before the improvement,indicating that the disease spots with lower temperature changes can be detected earlier.It facilitates earlier detection of disease spots for corresponding disease management.(5)Design and realize the identification of tomato disease detection GUI interface.GUI humancomputer interface design was carried out through Py Qt5 and Py Charm.Finally,on the basis of the generated tomato disease leaf disease detection model and algorithm,systematic debugging and experiments are carried out,and the results show that the designed tomato disease leaf detection interface is convenient to operate,simple and intuitive.This study proposes an early detection method for tomato diseases based on the characteristics of early stage diseases in certain crops.The method combines infrared thermal imaging for early detection and deep learning automatic recognition for identifying the tomato disease leaves.This approach enables early disease identification and improves the detection performance of tomato diseases,which can lay a foundation for early recognition of crop diseases.