Implementation of ResNet 101 Architecture in Convolutional Neural Network (CNN) Algorithm to Detect Diseases in Chili Plant Leaves Based on Image Processing

Abstract

Chili plants (Capsicum annuum L.) constitute a significant horticultural commodity in Indonesia, with production on the rise.  Chili producers frequently encounter difficulties, including insufficient knowledge regarding diseases that impact chili plant foliage and the application of technology.  Identified major diseases include leaf spot, curly top, and gemini, which can diminish chili output.  The application of the ResNet 101 architecture in Convolutional Neural Networks (CNN) for disease detection in plant pictures presents a viable approach.  The ResNet 101 architecture is employed to recognize and classify various illnesses on chili leaves, facilitating early symptom detection and diagnosis.  The residual architecture (ResNet 101) acquires intricate elements in images to improve disease diagnosis precision. This research examines the utilization, challenges, and benefits of disease detection systems through various methodologies, including observations and field investigations.  The process entails gathering picture models from various disease categories (leaf spot, curly top, gemini) and healthy leaves to construct a dataset, in addition to seeking expert consultation on illness classifications and employing the ResNet 101 architecture for modeling.  The use of the ResNet 101 architecture in the CNN model for disease detection on chili leaves, utilizing a dataset of 1,518 images categorized into four groups (leaf spot, curly top, gemini, and healthy leaves), yielded substantial outcomes, achieving an overall accuracy of 0.9382.  The established architecture must be evaluated against alternative designs to enhance the model's outcomes and to expand the dataset utilized to improve accuracy substantially.