Enhancing Plant Disease Identification Through Hyperspectral Imaging and AC-FNet Framework

The precise identification of plant diseases is essential for optimizing agricultural practices. The impact of these diseases on food production often leads to significant revenue loss. Detecting diseases that primarily manifest symptoms in plant leaves has conventionally depended on visual inspection by plant pathologists. However, modern methodologies use machine learning and computer vision techniques to facilitate disease identification, thus addressing the limitations of traditional approaches. However, in most of the existing works, the quality of the leaf image, smaller image regions, and chlorophyll content-based vegetation index features were not concentrated, which is required to be analyzed for the efficient detection of multiple leaf diseases. Therefore, this study highlights the necessity for a robust diagnostic system that tackles existing challenges in disease identification. Hyperspectral Images (HSI) serve as a valuable resource due to their rich spectral information. This research proposes a framework for multiple disease identification using hyperspectral data. The process begins by identifying target regions within the hyperspectral image through the Jeffries-Matusita-based Simple Linear Iterative Clustering (JM-SLIC) technique. Subsequently, the segmented image undergoes preprocessing, involving dead pixel replacement and noise removal. To approximate the resolution effectively, the Stochastic gradient-based Bi-cubic interpolation (S-BI) technique is employed. The resolution-approximated spectral images are then subjected to unmixing, followed by the estimation of chlorophyll content-based indexes. These indexes, combined with various features, contribute to disease identification within the leaf using the Atrous Convolution-based FractalNet (AC-FNet) model. The experimental outcomes strongly support the effectiveness of the proposed framework in detecting plant diseases efficiently.