Enhanced Detection of Bean Leaf Diseases Using a Stacked CNN Ensemble with Transfer Learning

Abstract

Bean leaf diseases are the major risk aspect for plant growth, and early detection is critical for farmers but challenging due to the complex structure of bean leaf diseases. Bean leaf diseases such as bean rust and angular leaf spots significantly diminish the quality and yield of agricultural products. Accurate detection is crucial for enhancing crop Yield and quality. To tackle this challenge, this paper proposes a novel approach using a deep stacked ensemble learning model, which combines the predictions of several pre-trained Convolutional Neural Network (CNN) models based on the Transfer Learning (TL) technique and utilizes a meta-learner on averaged predictions to detect bean leaf diseases. We have trained three pre-trained CNN models—EfficientNetB3, InceptionV3, and MobileNetV2—on a bean leaf dataset with 1296 leaf images and assessed their efficiency. Finally, we utilized a stacked ensemble learning approach, where the average of the predictions from these models are used as features to train an ensemble model to enhance the detection accuracy of bean leaf diseases. The proposed stacked ensemble method, particularly the combination of EfficientNetB3 and InceptionV3, achieves exceptional results with a classification accuracy of 99.22%, precision of 99.24%, recall of 99.22%, and F1-score of 99.22% on the test data with reduced training time, outperforming other state-of-the-art models.