CoffeeNet: A deep learning approach for coffee plant leaves diseases recognition

Abstract

Coffee is regarded as the highest consumed drink around the globe and has accounted as a major source of income in the regions where it is cultivated. To meet the coffee marketplace's requirements around the globe, cultivators must boost and analyze its cultivation and quality. Several factors like environmental changes and plant diseases are the major hindrance to increasing the yield of coffee. The development in the field of computer vision has facilitated the earliest diagnostic of diseased plant samples, however, the incidence of various image distortions i.e., color, light, size, orientation changes, and similarity in the healthy and diseased portions of examined samples are the major challenges in the effective recognition of various coffee plant leaf infections. The proposed work is focused to overwhelm the mentioned limitations by proposing a novel and effective DL model called the CoffeeNet. Explicitly, an improved CenterNet approach is proposed by introducing spatial-channel attention strategy-based ResNet-50 model for the computation of deep and disease-specific sample characteristics which are then classified by the 1-step detector of the CenterNet framework. We investigated the localization and cataloging outcomes of the suggested method on the Arabica coffee leaf repository which contains the images captured in the more realistic and complicated environmental constraints. The CoffeeNet model acquires a classification accuracy number of 98.54%, along with an mAP of 0.97 that is presenting the usefulness of our technique in localizing and categorizing various sorts of coffee plant leaf disorders.