<p>Crab image segmentation in underwater milieus represents major challenges due to poor visibility and complex backgrounds. Multi-level thresholding (MLT) is a widely used technique for partitioning an image into distinct classes, but conventional methods often struggle with accuracy and efficiency. This present study proposes an optimized segmentation model combining contrast-limited adaptive histogram equalization (CLAHE) for pre-processing and an improved Otsu method with a modified coati optimizer (MCO) for segmentation. CLAHE enhances image contrast, while the optimized Otsu method identifies suitable threshold values, reducing computational complexity. Each threshold value is treated as an element of a solution within the MCO, thereby improving threshold selection accuracy. The proposed model is evaluated using the sea animals image dataset and real-time video data. It achieved a Dice coefficient of 95.9% and an accuracy of 95.6%. Comparative analysis with traditional methods confirms superior segmentation accuracy and edge preservation. The model demonstrates enhanced robustness in detecting crabs under varying underwater conditions. The findings establish the proposed approach as a reliable solution for underwater image segmentation and crab identification.</p>