Enhancing Marine Biodiversity Monitoring through EfficientNetB5 for Sea Animal Image Classification

This research article describes a study on the categorization of photographs of marine creatures into 19 distinct groups based on their species. The study utilizes the EfficientNetB5 transfer learning model. The study employed the 'Sea Animals Image Dataset,' which comprised 11742 annotated images encompassing 19 species of sea animals, such as Coral, Crab, Dolphin, Eel, Jellyfish, Lobster, Nudibranch, Octopus, Penguin, Puffer, Sea Ray, Sea Urchin, Sea Horse, Seal, Shark, Squid, Starfish, Turtle/Tortoise, and Whale. The dataset was partitioned into three subsets for the purposes of training, validation, and testing, including 9393, 1174, and 1175 pictures, respectively. The EfficientNetB5 model was fine-tuned on the marine creatures dataset via transfer learning, using information acquired from the ImageNet dataset. The performance of the model was assessed by analyzing accuracy and loss graphs throughout the training and validation stages, and by examining a confusion matrix during the testing phase. The model attained a 91% accuracy rate on the test batch, showcasing its efficacy in categorizing marine animal photos. This study makes a valuable contribution to the area of marine biology and conservation by offering a dependable technique for automatically recognizing and classifying species of sea animals based on pictures. The categorization of marine organisms is crucial for the surveillance of oceanic ecosystems and comprehension of species variety, facilitating conservation endeavors targeted at safeguarding these crucial environments.