Gaussian filtering-based local ternary pattern for efficient classification of crop diseases / Modèle ternaire local basé sur le filtrage gaussienpour une classification efficace des maladies des cultures

Abstract

Accurate and reliable disease recognition in plants can assist in taking immediate remedial action, ad thus improve the overall productivity. In this work, we develop an intelligent machine-learning system accurately identify the diseases using leaf images of tomato plant. The images are represented in the re, saturation, value (HSV) format, and the V component is subjected to sub-band decomposition using aussian filters. Local ternary patterns (LTPs) are computed directly on the H and S components, and also 1 the decomposed images obtained from the $V$ component. The local texture information is augmented by obal information captured using histograms computed directly from the $\mathrm{H}, \mathrm{S}$ , and V components, to build comprehensive feature representation. The significant features are selected using the minimum redundancy aximum relevance (mRMR) algorithm and machine-learning techniques are applied for classification. The roposed feature identifies the various crop diseases more accurately than the existing methods.