Student Hamza Khaldoun Al-Khateeb discussed his master's thesis titled Recognizing Dangerous Camouflaged Aquatic Creatures Based on Machine Learning: Poisonous Stone Fish as a Case Study
Date : 16/12/2024
Supervised by Dr. Nawaf Omar Nawaf Al-Srehin.
Committee Members
Pro. Khalid Nahar Al-Nawasra
Dr. Mohammad Ahmad Alzoubi
Dr. Ra'afat Musa Alshorman
Abstract
Alkhateeb Hamza. K Recognizing Dangerous Camouflaged Aquatic Creatures Based on Machine Learning: Poisonous Stone Fish as a Case Study
Master's thesis, Yarmouk University (2024), (Supervisor: Dr. Nawaf Al-Sarihin) (Co-Supervisor: Dr. Khalid M.O. Nahar)
The toxic stonefish, scientifically known as Synanceia, is one of the most dangerous marine fish. These fish live mainly in shallow waters in the Indian and Pacific Oceans, where they are often found in sandy and gravelly areas near coral reefs. Because they have a high ability to camouflage themselves among rocks and coral reefs, their shape closely resembles rocks, making them very difficult to detect and posing a great danger to divers and swimmers who may encounter them without noticing their presence.
The study aimed to use a variety of machine learning and deep learning models, such as SVM, RF, and KNN, as well as CNN, ResNet50, and MobileNet, to achieve high performance in detecting stonefish. The performance was evaluated using model accuracy. Dreamtime, Saltwater, and Kaggle were the sources used to train the deep learning models, thus providing a wide and diverse set of data. Using precision, accuracy, recall, and F1 score metrics, the model performance was evaluated with accuracy and reliability. The results demonstrated the accuracy and success of the deep neural network method in detecting deadly stonefish. The fact that the deep learning models outperformed traditional machine learning models strengthened the confidence in its application for a range of tasks that require high accuracy in detection and identification. The CNN model outperformed the SVM model with an accuracy rate of 98.9%, while the SVM model achieved an accuracy rate of only 95.6%.
