Explainable techniques	Mechanism
Permutation Feature Importance (PFI)	The PFI is a technique for overall interpretability by examining the model score after shuffling a single feature value [31].
Local Interpretable Model-agnostic Explanation (LIME)	The LIME is a perturbation-based strategy that uses a surrogate interpretable model to substitute the complex model locally, providing local interpretability [31].
SHappley Additive exPlanation (SHAP)	The SHAP is a method for determining how each feature contributes to a specific outcome [31].
Faster Region with Convolutional Neural Network (R-CNN)	The faster R-CNN presented the Region Proposal Network [RPN], which speeds up the selective search. RPN adheres to the last convolution layer of CNN. Proposals from RPN are given to a region of interest pooling (RoI pooling), then classification and bounding-box regression [56].
Pseudo-coloring methodology	The pseudo-coloring methodology employs a range of colors to represent continuously changing values [29].
Class Activation Map (CAM)	The CAM uses global average pooling to generate class-specific heatmaps that indicate discriminative regions [30].
Value permutation and Feature-object semantics	The permutation of values is analyzed for their impact on predictions, and the most significant variables are then translated into statements using feature-object semantics [34].
Cumulative Fuzzy Class Membership Criterion (CFCMC)	The CFCMC offers a confidence measure for a test image's classification, followed by a representation of the training image and the most similar images [32].