Table 5.
Overview of XAI studies in plant phenotyping.
| References | XAI technique | Purpose of XAI | Phenotyping task | Plant dataset | Model |
|---|---|---|---|---|---|
| Ghosal et al. (2018) | Ranked features and generated saliency map of features | Explain model understanding | Identification of soybean stresses from plant leaves | Proposed dataset | Proposed model |
| Nagasubramanian et al. (2019) | Saliency map visualization | Track physiological insights of model prediction | Classification of charcoal rot | Proposed dataset | 3D CNN |
| Toda and Okura (2019) | 1. Occlusion analysis, 2. LIME, 3. GBP, 4. GradCAM, 5. DeepLIFT, 6. Explanation map | Interpret the representation of plant disease by a CNN | Plant disease classification | PlantVillage | InceptionV3 |
| Grinblat et al. (2016) | Saliency map visualization | Understand the features learned by a CNN for classification | Plant classification using vain morphological pattern of white bean, red bean, and soybean | Larese et al. (2014) | Shallow CNN |
| Wei et al. (2022) | 1. GradCAM, 2. LIME, 3. Smilkov et al. (2017) | Study the contribution of appearance and texture characteristics to model prediction | Leaf lesion classification | PlantVillage | 1. VGG, 2. GoogleNet, 3. ResNet |
| Mostafa et al. (2021) | GBP | Selection of model depth and analyzing overfit model | Plant and leaf classification | 1. PlantVillage | 1. Shallow CNN |
| 2. Plant Seedling | 2. ResNet50 | ||||
| 3. Beck et al., 2020 | |||||
| Mostafa et al. (2022) | GBP | Selection of model depth | Plant and leaf classification | 1. PlantVillage | 1. Shallow CNN |
| 2. Plant Seedling | 2. ResNet50 | ||||
| 3. Beck et al., 2020 | |||||
| Ghosal et al. (2017) | GradCAM | Isolate visual symptoms that contribute to model prediction | Classification of foliar stresses in the soybean plant | PlantVillage | Proposed model |
| Nagasubramanian et al. (2020) | 1. Saliency map, 2. SmoothGrad, 3. GBP. 4. Deep taylor decomposition, 5. Integrated gradients, 6. LRP, 7. Gradient times input | Compare different XAI techniques to interpret the prediction | Plant leaf classification | Ghosal et al., 2018 | DenseNet-121 |
| Minamikawa et al. (2022) | GradCAM | Visualize features relevant to the prediction | Measure the morphological features of citrus fruits | Proposed dataset | 1. VGG16 2. ResNet50 3. InceptionV3 4. InceptionResNetv2 |
| Akagi et al. (2020) | 1. GradCAM, 2. GBP, 3. LRP, 4. Guided GradCAM, 5. InceptionResNetv2 | Diagnose internal disorder in permission fruit using the visualization | Classify calyx-end cracking in persimmon fruit | Proposed dataset | 1. AlexNet 2. VGG16 3. InceptionV3 4. ResNet50 |
| Schramowski et al. (2020) | 1. GradCAM | Analyze Clever Hans-like behavior in deep learning models | HSI classification | 1. Proposed dataset | Proposed model |
| 2. LIME | 2. Fashion MNIST | ||||
| 3. Pascal VOC 2007 | |||||
| Desai et al. (2019) | GradCAM | Study image features that contribute toward the classification | Paddy rice's flowering panicle counter | Developed dataset | ResNet50 |
| Dobrescu et al. (2019) | 1. GBP | Study of the features extracted in regression | Count leaf of rosette plants | Leaf counting challenge Tsaftaris and Scharr (2017) | VGG16 |
| 2. LRP | |||||
| Lu et al. (2021) | Proposed visualization technique | Human interpretable visualization of the learned features of the proposed model | Count maize tassels, wheat ears, and rice plants | 1. Lu et al., 2017 | Proposed model |
| 2. Madec et al., 2019 | |||||
| 3. Liu et al., 2020 | |||||
| Drees et al. (2022) | Proposed visualization | Data augmentation | Data augmentation | Proposed dataset | Proposed model |