Table 2 Advantages and Limitations of Existing Work.
Methodology | Inference and Advantages | Limitations |
|---|---|---|
Pretrained CNN models | ||
Mask RCNN2,34, Sequential CNN4,6,26,28,38, DeepLab5, DeepLab-ResNet9,39, ResNet10, NAS-FPN20, Faster-RCNN20,25, DeepCNN21,30, MobileNet37, Multimodal CNN27, LGM based CNN31, ShuffleNetv2-Ghost32, Yolo multi-column deep NN40, Xception CNN36, Deep GAN41, VGG16 UNET42, Hybrid of CNN and GRU43, DCNN44 | The CNN models have the accuracy from 80 to 90% accuracy Model support portable data during the training Model achieves high generalization ability to perform the prediction Number of Epochs during training can be refined to improve the performance Limited data could be enough for better accuracy CNN model can be integrated with any other model for performance upgradation | Sometimes, the CNN model results with overfitting issues CNN need to perform better data preprocessing It also needs the input images to be processed with feature extraction methods CNN can provide high performance if the input images are processed with filtering methods It leads to have poor performance with raw data |
ANN based models | ||
Back-Propagation Artificial NN and chanvese active contour model3, Multi-layer NN24, | ANN models have the accuracy ranging from 80 to 90% ANN model automatically learns the complex patterns form e input dataset images | ANN models perform well only with labelled data Normalization of input dataset is essential for high performance |
Image processing based models | ||
FC estimation algorithm included HSL thresholding method7, WS and CHT algorithms8, SIFT, HoG22, Color constancy method23, GLCM29, Correlated color discrepancy33, LBP LVP35 | This model is so simple and easy to implement Customization of images can be done well to perform image filtration Less data is enough to perform well Computational requirements are low | Image processing models need images with high resolution condition Fitting the pixel threshold value is difficult to proa challenging task Model does not learn the contextual information of the image |
ML based models | ||
Performance of the ML model can be increased by fine tuning the model parameters | Model be contingent on the pre-defined structures in the images | |
