<p>The underwater environment is recognized as a particularly complex setting, requiring accurate detection techniques to ensure continuous and effective monitoring of fish species. However, traditional computer vision methods have limitations when faced with environmental constraints such as water turbidity, the natural camouflage of fish, degraded image quality, and the high similarity between objects of interest and the background. This paper presents a new deep learning architecture, named BiOLO-Wave, dedicated to the detection and classification of fish in underwater environments. The developed approach ensures high robustness under degraded visual conditions. It also stands out for its ability to produce localizations that align with fish morphology, even in the presence of various orientations or postures. The BiOLO-Wave architecture is based on a YOLOv11 model, enhanced by several innovative modules. A Chromatic-Wavelet Attention (C-Wave) module is integrated upstream of the backbone to ensure an efficient transition between the spatial and frequency domains. Next, three Dynamic Gated Adaptive Attention (DGAA) blocks are integrated into the neck, acting as adaptive attention filters to dynamically recalibrate information flows based on the local environmental context. Finally, an Oriented Deformable Convolution (ODC-Head) module is introduced to replace traditional prediction heads with a directed deformable convolution structure. This mechanism enables the prediction of oriented bounding boxes (OBBs). Experimental results demonstrate the effectiveness of the proposed approach, with an accuracy of 93.88%, a mAP@0.5 of 88.90%, and an inference latency of 12.54, thereby confirming the relevance of BiOLO-Wave for real-time aquaculture monitoring applications, particularly in turbid underwater environments.</p>