MAFH Based Maize Leaf Disease Detection and Classification

Abstract

Abstract Maize plant plays crucial role not only in the field of agriculture but also in global economy, since it is the third most cultivated crop across the globe. However, these plants are usually affected by various types of diseases such as blight, common rust, gray leaf spot, etc., protecting the plants from these disease is very important. This research proposes a new deep learning model for disease detection to perform well than other models. The hybrid model uses MobileNetV3 as the backbone architecture integrated with attention, fusion and head. The framework includes data preprocessing, feature extraction, classification and interpretation. We will compare the performance of this model with other models such as VGG16, ResNet50, DenseNet121, ALEXNET, etc. criteria for the final evaluation includes Accuracy, Precision, Recall, F1score, Specificity, Logloss, AUC-ROC curve. Through this proposed model we have achieved an accuracy of 98% which is high than the other models compared. The lightweight nature of MobileNetV3 enables us to implement the model in the mobile and IoT devices also. The present study contributes to the development of deep learning model in the field of agriculture, offering a efficient solution for early maize leaf disease detection.

Plant phenotyping relevance

トウモロコシ葉の病徴を画像から検出・分類する深層学習手法の開発と比較が中心であり、植物病害状態の画像ベース表現型計測に該当する。

Code and data availability

The paper's maize leaf disease detection model (MAFH) was trained and evaluated entirely on a public Kaggle image dataset, which the authors explicitly declare in the Data Availability statement. No author code, trained model checkpoints, or other paper-specific assets are stated as publicly available.