Table 1 Summary of research works.
Authors | Methodology/algorithm | Merits | Limitations |
|---|---|---|---|
Chougala & Ramachandra, 2020 | Color histogram + K-Means + Edge Detection | Disease intensity estimation through area-based calculations | Accuracy drops under varied lighting conditions |
Umamaheswari et al., 2022 | VGG-16-based classification model | Effective in disease classification; offers visual analysis and treatment recommendations | High resource usage; not optimized for complex symptoms without tuning |
Devisurya et al., 2022 | Enhanced YOLOv3-Tiny + Residual Network | Enhanced detection accuracy using light architecture; better than Faster R-CNN | Accuracy affected by limited data scale; deeper networks might overfit on small datasets |
Balafas et al., 2023 | Disease detection using YOLOv5 | Demonstrated YOLOv5’s advantage in object detection; better efficiency | Failed to address noisy data handling and early disease identification in dynamic field conditions |
Kannan & Thangavel, 2023 | RGB-to-HSI conversion + SFCM clustering + LDA-ANFIS classifier | High classification accuracy using hybrid fuzzy inference; enhanced feature extraction | Poor generalization due to small dataset; computational load increases with added stages |
Hosny et al., 2023 | Deep CNN + LBP (Local Binary Pattern) | Captures both spatial and textural features; improved accuracy over standard CNNs | Integration of multiple features increases training complexity |
Shrotriya et al., 2024 | Image segmentation + SVM and K-Means | Improved detection by isolating image regions and using multiple descriptors | Performance is dataset-dependent; requires extensive training data for consistency |
Patil et al., 2024 | CNN feature extraction + SVM classifier | Strong accuracy and faster prediction; suited for real-time systems | Prone to overfitting due to fewer training iterations |
Moupojou et al., 2024 | SAM segmentation + FCDD for classification | Handles real-world field images; segments leave even with complex backgrounds | Misclassifies areas with dense foliage; needs enhancement for green-dominant images |
Madhurya & Jubilson, 2024 | YOLOv7 + CLAHE + ShuffleNetV2 + ERSO + RFO | Combines multiple models for precise detection; good contrast enhancement | High computational cost due to layered architecture |
Rashid et al., 2024 | IoT integration with CNN in MMF-Net | Multi-contextual fusion (image + environment); suitable for real-time monitoring | Relies on IoT network stability; environmental variability may affect predictions |
Zhang et al., 2024 | Modified ResNet + Capsule Network | Preserves feature location and captures fine textures; uses attention mechanisms | High model complexity; inference time unsuitable for edge devices |
Paramanandham et al., 2024 | LeafNet + Residual Learning + Xavier initialization | Robust in noisy conditions; dual loss functions refine training | Time-intensive training; lacks transferability to different crops |
