Enhancing marine species detection and classification with a hybrid GCS-Yolov5-Hcnn model for underwater monitoring

Detecting and classifying marine species in underwater environments presents challenges such as low visibility, varying object sizes, and the need for accurate species identification. Existing methods often struggle with these complexities. This research introduces a novel hybrid approach, named Genetic-based Cuckoo Search Algorithm-You Only Look Once V5 with Hierarchical CNN (GCS-YV5bHCNN), to address these issues and enhance the accuracy and reliability of marine species detection and classification in underwater marine images. The methodology begins with pre-processing marine species image datasets to standardise and improve image quality. It employs the GCS optimisation technique to iteratively train YOLOv5 and Hierarchical CNN models, refining their parameters for superior performance. YOLOv5 is utilised for efficient object detection, pinpointing, and locating marine species, while Hierarchical Convolutional Neural Network (HCNN) classifies these detected objects into specific species categories. The system facilitates ongoing monitoring of marine populations and environmental conditions, offering valuable insights into ecological changes. The method is thoroughly assessed using metrics such as precision, accuracy, recall, mean average precision (mAP), and detection time, demonstrating significant enhancements in marine environment monitoring. Finally, the GCS-YV5bHCNN model achieved an accuracy of 99.64%, demonstrating its superior performance in accurately identifying and classifying images compared to the other methods.