An effective feature descriptor method to classify plant leaf diseases using eXtreme Gradient Boost

Abstract

Identifying plant leaf diseases will be highly difficult due to the difficulties in gathering lesion characteristics from a quickly changing atmosphere, imbalanced illumination reflection of the incoming light source, and numerous other factors. A practical strategy for classifying plant leaf diseases is provided in this research. Using HSV, HU moments, and color histograms, we first created a leaf feature improvement framework that can enhance leaf characteristics in a complicated environment. Then, to increase feature classification capacity, a competent extreme boost method is modelled. Batch normalization is used to avoid network overfitting while also improving the model's resilience. The plant leaf disease feature improvement approach is favorable to boosting the efficiency of the XGBoost classification, as demonstrated in studies from various perspectives. For plant leaf disease photos obtained in the natural environment, our technique displays significant resilience, serving as a benchmark for the intelligent categorization of additional plant leaf diseases.