Neem leaf disease detection from near and far field images using deep learning techniques

Abstract

In India, Neem plant plays an important role in the manufacture of various herbal products for humans and animals. Neem tree growth, and leaves are affected by biotic and abiotic agents. Biotic agents are bacteria, fungus, and viruses. Abiotic agents are water, temperature, and humidity. Biotic active ingredients reduce plant growth and reduce the production of high-quality herbal products. Neem leaves are affected by bacteria. Neem leaf diseases look similar in texture, shape, and color. So, farmers never easily recognize symptoms of neem leaf disease. Pathogens can recognize neem leaf diseases, through the pattern, size, shape, and texture of the leaves. Neem leaf diseases are identified through laboratory tests such as the polymerase chain reaction (PCR) test. PCR is more expensive, and time consuming. In this thesis, to avoid the above problem, neem leaf diseases are analyzed using deep learning techniques such as LCFN, FBFN, and TL. Neem leaf disease features are extracted with a leaky capacitor-fired neuron (LCFN) model, the model extracts 1D time sequence from the leaf image. LCFN sequence includes the texture of the affected leaf area, outline, and the area of the leaf lesion. The LCFN model creates a continuous sequence of images by combining pixels from neighboring images. LCFN features are combined with the morphological features of the leaf lesions, and applied with classifiers. The ensemble classifier provides a recognition accuracy in neem leaf diseases of 82.2%. Next, fuzzy-based model of the fired neuron (FBFN) extracts the optimized features of neem leaves by optimizing the parameters of fired neuron and results are passed through the adaptive neuro fuzzy classifier and the model provides an accuracy in neem leaf diseases of 94.4%. Next newline