International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

A Dual-Branch Attention-Enhanced CNN for Corn Leaf Disease Classification via RGB-HLS Color Space Fusion

Authors :

Mohammad Ali Salehi Rad¹ Kamran Kazemi² Mohammad Sadegh Helfroush³ Tahereh Golshaeian⁴

1- Department of Electrical Engineering Shiraz University of Technology 2- Department of Electrical Engineering Shiraz University of Technology 3- Department of Electrical Engineering Shiraz University of Technology 4- Guilan Agricultural Organization

Keywords :

Plant disease classification،Convolutional neural network (CNN)،Color space،DenseNet،Convolutional block attention module (CBAM)

Abstract :

Accurate classification of plant diseases, achieved through machine learning methods such as convolutional neural networks (CNNs), is essential for improving crop productivity and reducing agricultural losses. However, most studies have only used RGB images as input. Incorporating multiple color spaces simultaneously can capture complementary spectral characteristics and improve classification accuracy. In this study, we proposed a dual-branch DenseNet121 model for classifying corn leaf diseases. The model processed images in both RGB and HLS color spaces separately. Each branch extracted features independently, and a convolutional block attention module (CBAM) was used at the end of each branch to help the network focus on important regions of the image. This dual-branch design exploited the complementary strengths of both RGB and HLS color spaces by combining their features, providing the model with more comprehensive and discriminative representations to improve disease classification. Evaluation results on the PlantVillage dataset showed an accuracy of 98.38%, outperforming other CNN-based models. These findings demonstrated that integrating multiple color spaces with attention mechanisms was an effective approach for plant disease detection.