Precision Agriculture-Oriented Cauliflower Foliar Disease Detection Using Efficient Net-B3

Cauliflower cultivation plays a vital role in global vegetable production; however, foliar diseases significantly reduce crop yield, quality, and commercial value. Conventional disease diagnosis methods largely depend on manual inspection by agricultural experts, which is labor-intensive, time-consuming, and often inconsistent under large-scale farming conditions. This research proposes a deep learning-based automated disease recognition framework utilizing a fine-tuned EfficientNet-B3 architecture for accurate classification of cauliflower leaf diseases. The proposed framework categorizes cauliflower leaves into four classes: bacterial spot, black rot, downy mildew, and healthy leaves. A dataset containing 650 labeled leaf images collected from field environments and agricultural repositories was employed for experimentation. Transfer learning was adopted using ImageNet pre-trained weights, followed by fine-tuning of higher convolutional layers to adapt the network to disease-specific visual characteristics. Extensive preprocessing and augmentation techniques were incorporated to enhance robustness against illumination variation, leaf orientation, and background noise. Experimental evaluation demonstrates that the proposed model achieves an overall classification accuracy of 98.48%, along with excellent precision, recall, and F1-score across all classes. The findings validate the effectiveness of EfficientNet-B3 in identifying subtle disease patterns and demonstrate its applicability for real-time precision agriculture systems. The proposed approach can support farmers in early disease diagnosis, optimized pesticide utilization, and sustainable crop management.