Major cotton leaf diseases include Bacterial Blight, Curl Virus and Fusarium Wilt and cause significant reduction in crop productivity. Deep learning architectures such as VGG16 and VGG19 have been used for automated disease detection. Using transfer learning and data augmentation, both models were fine-tuned on a dataset comprising four classes: Bacterial Blight, Curl Virus, Fusarium Wilt, and Healthy. Results show the validation accuracy of VGG16 was higher, 91.13 versus 83.16% for VGG19. Its robust feature extraction and generalization capability was also demonstrated by VGG16 in terms of precision (85.60%), recall (85.22%), F1-score (85.13%). The training accuracy of both models was very high (> 99%), and VGG16 was more reliable for real world usage. The outcomes of this learning make clear the efficiency of VGG16 for disease detection in precision agriculture, as a scalable early diagnosis and timely intervention solution. Future work will expand the dataset, optimize it further, and deploy the system on mobile devices in real time for field use.