Deep CNN and Deep GAN in Computational Visual Perception-Driven Image Analysis

Table 8

Challenges and open problems for computational visual perception-driven image analysis.

Research direction	Key references	Dataset	Challenges and open problems	Future research
Water level measurement	[132, 133]	GNSS dataset	Image-based in situ water level measurement faces several challenges: image distortions, low visibility, and ambient noises.	The image can be changed from 2D to 3D, and the water level measurement can be done by utilizing advanced computer vision techniques.
Image processing	[134, 135]	Sequential image captured with the camera	Classical image processing techniques are based on a controlled environment. Environmental changes require image processing techniques such as custom filters, thresholding, and limited site coverage.	The model is generalized using deep learning techniques in a dynamic environment.
Flood detection	[136]	Manually collected dataset	Lack of open-source data to train computer vision algorithms.	The data can be collected and opened to train the proposed model.
UAV image processing	[137, 138]	Datasets of AOGCM simulations	Proposed solutions have limited generalizability.	Advanced convolutional neural networks can be used. Instead of using image processing techniques, the model’s generalizability can be assessed using real-world data for the testing phase.