Publication

Abstract : Research into marine environments and efforts to preserve marine life rely heavily on Underwater Object Detection (UOD). Water turbidity and changing lighting conditions are two of the environmental variables that make underwater item recognition and classification quite difficult. Limited underwater image dataset is also a limitation in training the model. Generative Adversarial Networks (GANs) are used in this research that mainly focuses on creating data that is similar to current data. The Multi-Feature Attention CycleGAN (MFA-CycleGAN) model is used in this research with extending its features. Multi Level convolutional and residual layers are added to the generator and discriminator networks so that the model gains a greater capacity to learn complex and fine-grained features that are characteristic of underwater imagery. Squeeze-and-Excitation (SE) blocks are integrated to this model to consider informative spatial and channel-wise features. In order to perform the object detection, YOLOV8 is used. To improve the ability to recognize small or concealed underwater objects, dual Convolutional and residual layers are added for the acquisition of more spatial and contextual information. Route aggregation blocks are added to the neck section improves the model’s ability to fuse features from different scales, which in turn improves its ability to understand objects at different resolutions and sizes. The proposed Enhanced Multi-Feature Attention CycleGAN with Enhanced YOLOV8(EMFA-CycleGAN-EYOLOV8) accurately detects the underwater images. The proposed model when compared with the traditional models performs better in underwater object detection. The proposed model achieved 96.8% accuracy in Image generation using EMFA-CycleGAN, 95.2% accuracy in YOLOv8 feature processing and 97.5% accuracy in object detection.