Plant Disease Detection Using Enhanced EfficientNet Architecture in Comparison with DenseNet to Analyze the Severity in Leaves with Performance Measures

Abstract

The backbone of the Indian country is enriched with growing crops and cultivation in an effective way. The economical background is a major impact on countries’ financial stages where there is a lack of knowledge to support agriculture as well. In accordance with agriculture and plants development assessment of needs are also analyzed. However, there are several problems to be addressed in crops and plant diseases and the main focus is to reduce the disease on leaves texture that can be associated with the health and wealth of the country. To detect plant diseases there are several methods applied from existing techniques such as Machine Learning (ML) and Deep Learning (DL) where the species produces various results and comparison of results are facing certain limitations. To overcome the limitations in the existing deep learning Convolution Neural Network (CNN) framework such as DenseNet, the Enhanced EfficientNet (EEN) is proposed in the research work to overcome the limitations. To predict the plant disease in the early phase using the proposed algorithm can produce better accuracy such that the quality of crops can be protected without diseases affected. The main step in the proposed idea is to examine the performance by detecting the diseases in leaves by analyzing images from the dataset downloaded from Kaggle in the name of PlantVillage. These datasets were used with the leaf images to detect the severity of plants by extracting multiple features such as textures of multi-labeled leaves, color deviation in the early stage, and shape variations also considered. Classifier using EEN preprocess and classifies accordingly. There is an elimination of regions after segmenting the image according to the classes. Such feature mapping gives the exact results faster with 99.8% of accuracy.