Research Progress in Image-based Plant Disease Identification
Plant disease identification is an essential link and prerequisite in plant disease pre-vention and control,and the development of high-precision disease identification technology has be-come an urgent requirement in efficient disease prevention and control.Plant disease image identifi-cation research began in the 1980s,which plays a crucial role in agricultural production and plant protection.Timely and accurately identification of plant diseases can assist in distinguishing disease types and implementing corresponding preventive measures,thereby alleviating the adverse effects of diseases on crop yield and quality.Based on the compilation and summary of existing domestic and international research literature,this article focused on plant disease image identification technol-ogy from three aspects:image segmentation,feature extraction,and classification identification.It also provided an outlook on the future development of plant disease identification.At present,the methods of image segmentation of plant diseases are mainly based on threshold,clustering,edge and deep learning techniques,which can be divided into fixed threshold method and adaptive threshold method.The fixed threshold method is to manually try different values based on the difference be-tween the target and the background pixel histogram,and select the appropriate threshold value to achieve image segmentation.Adaptive threshold method is based on the principle of fixed threshold segmentation.According to specific rules,the threshold is obtained by automatic iteration of com-puter.The most common one is the maximum inter-class variance method(Otsu method),while the segmentation effect of the threshold segmentation method based on GCLPSO is better than that of other similar algorithms,with good convergence and stability.In the plant disease identification re-search,feature extraction and classification identification are the key factors affecting the identifica-tion rate.Feature extraction is to describe the attributes,obtain the disease information,and find out the most useful discrimination features.Classification recognition is based on image segmentation and feature extraction,through the construction of classifiers.The future development direction of in-telligent plant protection is to apply the methods of transfer learning and lightweight network to plant disease identification,and to study and design a network model with high identification accu-racy.Image-based plant disease identification could offer more scientific and intelligent support for disease monitoring and prevention,holding significant importance for global food production and sustainable agricultural development.
国家科技期刊平台
NSTL
万方数据
