Enhancing Chili Plant Disease Detection: A Comprehensive Ensemble Approach Utilizing Pre-Trained Deep Learning Models

Chili is one of the most prominent crops of Bangladesh. Chilis are great at boosting the human immune system and contributing to overall good health. But every year, chili production is hampered due to various diseases. Deep learning-based models can mitigate this issue by detecting plant diseases quickly and accurately. This research aims to make an ensemble technique approach harnessing various pre-trained CNN models to classify healthy chili leaves and chili diseases such as yellowish, leaf spot (fungi), and whitefly infestations. We used CNN models such as MobileNetV2, ResNet152, EfficientNetB7, EfficientNetV2B0, InceptionV3, ResNet50, DenseNet121, and DenseNet201. The Chili dataset was approved by BARI (Bangladesh Agricultural Research Institute). We examined the outcomes of each model with three optimizers, Adam, RMSRrop, and SGD, where RMSProp gave the best outcome overall. Then, ensemble techniques were implemented by combining two variations for each of the top three best-performed CNN models, subsequently comparing their performance. The proposed ensemble model is EfficientNetV2B0, with EfficientNetB7 outperforming all the existing deep learning models and achieving a maximum accuracy of 99.56%.