• Developed a deep learning pipeline for tomato leaf disease classification using three state-of-the-art CNN architectures—VGG19, ResNet152, and EfficientNetB7—fine-tuned with transfer learning on a 10-class dataset, including real-time field images from Palghar, Maharashtra.
• Benchmarked each architecture across three optimizers (Adam, SGD, RMSProp), uncovering optimizer-dependent performance patterns and identifying RMSProp as the most consistent performer overall.
• Engineered a novel hybrid ensemble model combining all three CNNs, fused via softmax probability averaging and enhanced with dropout and L2 regularization to ensure robustness and generalization.
• Achieved 99.2% test accuracy, outperforming individual architectures and validating the model's effectiveness on real-world agricultural data, including unseen field images.
• Addressed a significant research gap in optimizer-aware hybrid model design for plant disease classification, enabling scalable AI deployment in precision agriculture systems.

Venue: 5th ICIVC 2025; to appear in Springer's Lecture Notes in Networks and Systems, indexed in Scopus

• Designed and implemented a hybrid spatio-temporal deep learning framework integrating CNN for spatial feature extraction and LSTM networks for temporal modeling to classify plant leaf diseases and quantify their severity on a 5-level scale across time-sequenced images.
• Utilized CycleGAN-generated synthetic sequences alongside real-time images from Palghar, Maharashtra, to simulate disease progression and address data scarcity in intermediate severity stages.
• Integrated temporal modeling to capture visual symptom evolution over time, enabling early-stage detection critical for timely interventions in smart agriculture.
• Achieved over 97% classification accuracy across four crops—tomato, apple, bell pepper, and cherry—demonstrating strong generalization and model scalability across diverse agricultural scenarios.
• Validated model robustness under field conditions with varying lighting, occlusion, and background noise, supporting deployment in real-time plant health monitoring systems.

Venue: 16th ICCCNT 2025; to be published in IEEE Xplore and indexed in Scopus