COLEN-An Improvised Deep Learning Model for Plant Disease Detection Using Varying Color Space

Abstract

Plant disease detection is a rapidly evolving area of research with new techniques being discovered and proposed almost every year. Most work done in this field focuses upon increasing the accuracy of the classifiers, however when it comes to real images taken from fields, the accuracy drops drastically due to varying light intensity and background of the images. In this paper a novice approach has been used to improvise the performance of the plant disease detection model. It has been seen from the tremendous experimentation that, mostly in Red Green Blue (RGB) input images, while detecting the diseased leaf images, the green channel in healthy part dominates the blue channel in the diseased part of the same leaf which is found to be one of the other factors which is reducing the accuracy. Hence, here eight widely used color spaces are considered to measure the impact of the performance of the model by changing the color space. For conducting the experimentation on the proposed idea, a public dataset is being used in varying color spaces against RGB space. From the experimentation results it has been observed that for the majority of the classifiers, for the Hue Saturation and Lightness (HSL) color space the accuracy was more than the accuracy for RGB or any other color space. Furthermore, to improvise the performance, an ensemble algorithm named as COLEN has been proposed to choose the color space with the maximum accuracy and F1 score.