| The apple industry is one of the most vital parts of China’s fruit industry,and over the years,China has become one of the world’s largest apple producing and consuming countries.The growth of the apple industry not only satisfies people’s nutritional needs,but also brings huge benefits to local economic development.However,due to environmental factors and the influence of pathogens,apples are prone to a variety of diseases leading to reduced yields,lower quality and ultimately economic losses.Therefore,the study and prevention of apple diseases is of great importance to ensure the sustainable development of the apple industry.Traditional methods of control based on experience are not only ineffective but may also cause environmental pollution,so it is essential to use modern computer technology to detect and identify apple diseases.To address the problems in existing research on apple disease detection and identification,this study investigates three common apple leaf diseases,selects YOLOX-Nano as the basic network framework to train apple disease detection models,carries out network optimisation to improve model performance,balances real-time and accuracy of the models,and designs and implements a mobile-based apple disease detection application with the following contributions:(1)Construction of apple leaf disease dataset in complex backgrounds.In order to improve the practicality of the model and to achieve the detection and recognition of apple diseases in complex scenes,this study uses both offline and online data collection to construct a dataset of apple leaf diseases in natural scenes.The offline data was collected from leaf images in apple orchards and forests,while the online data was collated from existing public datasets,followed by data annotation using tools.The dataset contains images of three types of diseased leaves in complex contexts,enabling the model to better adapt to the effects of different lighting,shading,angles and image quality in real environments,and to meet the detection needs of practical use.(2)Research on lightweight apple leaf disease detection methods.In this study,a lightweight and efficient leaf disease detection model is constructed based on the YOLOX-Nano network method.Firstly,the bottleneck blocks in the backbone network are optimised,the module performance is improved using feature reuse,and asymmetric shuffleblock are proposed to enhance the network’s ability to acquire disease features.Then,a lightweight attention mechanism is embedded in the CSP module of the network to rescale the feature weights,allowing the network to focus more on information related to disease classification during the learning process,reducing the interference of background factors and improving the detection accuracy of the model.The Depthwise Separable Convolution in the network is replaced with a betterperforming Blueprint Separable Convolution to ensure that the model is lightweight while improving model performance.In addition,the use of CIo U-Loss in the network for regression prediction allows the model to regress more accurately on the disease target.Comparing with other methods on the self-built dataset MSALDD and the public plant disease dataset Plant Doc,the results show that the method proposed in this study can quickly and accurately detect apple leaf diseases in complex environments with the highest detection accuracy and fast detection speed,proving the generalisability and effectiveness of the method.(3)Software development for apple leaf disease detection.To expand the practical application scenarios of the model,the disease detection model is deployed for mobile use in this paper,and the apple leaf disease detection software is designed and developed.The application can detect diseased leaf targets in pictures and real-time videos,and display the detection results to the user with feedback on disease information and control methods,helping the user to understand the details of the disease and enhancing the usefulness of the model in this paper. |
PDF Full Text Request