Table 1 Evaluation of the proposed model against existing methods.
S. No | Existing Approaches | Objectives | Dataset Quantity | Model Used | Accuracy | Model Size/Inference Characteristics |
|---|---|---|---|---|---|---|
1 | Wang et al.46 | Classification of common plant leaf diseases using conventional CNN Architecture | 10,000 images | VGGNet | 95.45% | Large model, high parameter count |
2 | Chen et al.47 | Ensemble learning for enhanced feature representation for plant disease detection | 12,000 images | Ensemble CNNs | 96.20% | High computational overhead |
3 | Liu et al.48 | Transfer learning for accurate plant disease identification | 13,000 images | EfficientNet-B0 | 97.80% | Compact, efficient |
4 | Lee et al.49 | Adapting a real-time object detection framework for plant disease identification | 15,000 images | YOLOv3 | 94.70% | Fast inference, detection-oriented |
5 | Ahmed et al.50 | Classifying Plant disease using hand crafted features with traditional ML | 8,000 images | CNN + SVM/RF | 89.50%(SVM), 87.90 (RF) | Low inference speed, feature-heavy |
6 | Sharma et al.51 | Improving gradient flow and feature reuse for enhanced disease classification | 20,000 images | DenseNet-121 | 96.75% | Moderate–high parameters |
7 | Patel et al.52 | Designing a light weight model suitable for edge device deployment | 18,500 images | MobileNetV2 | 95.20% | Lightweight, fast inference |
8 | Singh et al.53 | Efficient multi-scale feature extraction for plant disease classification | 15,000 images | InceptionV3 | 97.10% | Moderate computational cost |
9 | Roy et al.54 | Integrating temporal features with CNN extraction for plant disease diagnosis | 12,000 images | Custom Hybrid CNN-RNN | 96.00% | Increased temporal complexity |
10 | Proposed Work | Hybrid approach combining local and global feature extraction for multi class plant disease diagnosis | 21,534 images | EfficientNet-B7 + ViT-B16 | 98.13% | High-capacity hybrid, optimized feature reuse and improved accuracy with acceptable inference cost |
