Table 1 Literature survey.
Author(s) | Year | Publication | Methodology | Accuracy | Advantages | Limitations | Research Gap |
|---|---|---|---|---|---|---|---|
Sankar Sennan et al22 | 2022 | Computers, Materials & Continua | Custom CNN on four leaf types | 97.50% | High accuracy; compared multiple baselines | Small (~ 400 images); non-spinach types | No XAI; limited generalization |
Yıldırım and Yalçın23 | 2024 | J. Food Nutr. Res | ResNet-101-based CNN for spinach freshness | ≥ 89.4% | Good baseline for food quality tasks | Limited accuracy; no interpretability | Absence of XAI; no hybrid ensemble |
He et al24 | 2024 | Infrared Physics & Technology | Hyperspectral + DL classifiers (spinach + cabbage) | > 80% | Non-destructive biochemical analysis | Expensive instrumentation; lower accuracy | No image-CNN/XAI; equipment-heavy |
Kumar et al.25 | 2024 | Current Research in Food Science | CNN–BiLSTM hybrid for generic vegetable freshness | 97.76% | Models spatial & temporal features | Computationally heavy; generic to veggies | Not spinach-specific; lacks visual explanations |
Tapia-Mendez et al26 | 2023 | Applied Sciences | MobileNetV2 ensemble for fruit & vegetable ripeness | 97.86% | High accuracy across ripeness stages | Broad domain; no spinach focus; no XAI | No spinach dataset; no explainability |
Yuan & Chen27 | 2024 | Current Research in Food Science | Deep feature fusion (GoogLeNet, DenseNet-201, ResNeXt-101) + PCA + SVM | 96.98% | No CNN retraining; efficient feature-based detection | Not spinach-specific; image quality threshold unclear | No explainable visualization |
Koyama et al.28 | 2021 | PLOS ONE | Color & local feature + SVM/ANN for spinach freshness | 84% (2-class) | Non-destructive; validated against sensory panel | Lower accuracy; no deep learning | No image-based DL; no XAI; training on small smartphone dataset |
Elumalai and Meganathan29 | 2024 | J. Robotics & Control | Hybrid of Orange-embedded pre-trained models + ML classifiers on spinach leaves | ~ 99% | High accuracy; variety classification | Limited information on freshness; tool-specific | No interpretability; no confidence-based recommendation logic |
