International Journal of Computer Applications |
Foundation of Computer Science (FCS), NY, USA |
Volume 186 - Number 67 |
Year of Publication: 2025 |
Authors: Md Akash Rahman, Md. Safi Ullah, Rimon Kanthi Devnath, Taufiqul Hoque Chowdhury, Gulapur Rahman, Md Atikur Rahman |
![]() |
Md Akash Rahman, Md. Safi Ullah, Rimon Kanthi Devnath, Taufiqul Hoque Chowdhury, Gulapur Rahman, Md Atikur Rahman . Cotton Leaf Disease Detection: An Integration of CBAM with Deep Learning Approaches. International Journal of Computer Applications. 186, 67 ( Feb 2025), 1-8. DOI=10.5120/ijca2025924487
Cotton is a major contributor to Bangladesh’s economy, serving as one of the primary cash crops. However, cotton production faces substantial challenges due to various diseases affecting the leaves, collectively referred to as Cotton Leaf Disease. Bacterial blight, leaf curl virus, fungal infections, and pest attacks negatively impact crop yield and quality, leading to economic losses. Conventional manual inspection techniques are inefficient, labor intensive, and prone to inaccuracies, which hinder timely disease identification and intervention. This study presents a deep learning system incorporating the Convolutional Block Attention Module (CBAM) for the automatic detection of cotton leaf diseases. A public dataset was utilized, consisting of 2,137 images in the original set and 7,000 images in the augmented set, categorized into seven classes: Bacterial Blight, Curl Virus, Herbicide Growth Damage, Leaf Hopper Jassids, Leaf Reddening, Leaf Variegation, and Healthy Leaf. Multiple deep learning models, including EfficientNetB1, DenseNet121, DenseNet169, MobileNet, Xception, and InceptionV3, were trained following the application of critical preprocessing techniques such as image resizing, noise reduction, and image normalization to improve image quality. The integration of CBAM into the models enhanced the emphasis on relevant image features, thereby improving detection performance. Among the models evaluated, DenseNet169 achieved an accuracy of 96.26% on the original dataset, whereas EfficientNetB1 with CBAM attained the highest accuracy of 99.21% on the augmented dataset.