Early Symptom Plant Leaf Pathology Recognition via Hybrid Spatiotemporal Deep Intelligence Framework
The timely appearance of plant leaves diseases is critical to averting a loss of crop yields and save the use of unnecessary chemical pesticides. The common visual inspection techniques used are not suitable in detecting the diseases at early stages of the symptoms, thus intervention is not timely. This paper shows an intelligent plant early disease identification system at the early development stage that uses a combination of image and deep-learning technology. Images of the leaves that have been taken in the natural field conditions are preprocessed using the noise reduction, color normalization and lesion enhancement algorithms, which are implemented in MATLAB, to make the features more visible. A deep belief net, based on a hybrid architecture, comprising of Long Short-Term Memory forms the basis of extracting strong spatial and temporal features to classify accurately healthy and disease leaves with the first level of symptoms. The system also assesses the extent of the disease and suggests measures to take to treat it, thus focusing on organic and biological measures of mild cases of infection and only suggesting chemical treatment when needed. Experiment reports prove better early detection, lesser chemical addiction, and greater environmental sustainability and data-driven agricultural disease management practices.