Metrics Evaluation of Bell Pepper Disease Classification Using Deep Convolutional Neural Network (DCNN)

Crop disease is one of the major issues in agricultural fields. It reduces the food production process and produces a huge economic loss for farmers in agricultural lands. Deep learning models are built for detecting and identifying diseases at an early stage to help the farmers and provide food security. In this paper, transfer learning approach is used to identify and classify the diseased portion of leaves in farming lands. The benchmark dataset was taken from the Internet (Kaggle) for bell pepper leaf for identifying the diseases such as pepper leaf spot, mosaic virus, blight, damping off disease, verticillium wilt, blossom end rot, yellowing leaves, curling leaves, and healthy leaves. Ten pre-trained models are built for recognizing diseases and compared to obtain better accuracy. Adam optimizer and binary cross-entropy is used in the model for calculating loss function. In this paper, experimental results are presented on bell pepper diseases for MobileNetV2 with better accuracy of 99.42%. The various performance metrics such as accuracy, precision, recall and F1-score, ROC curve are used to determine the accuracy of the model.