Shrimp vision: Improved shrimp detection using underwater image enhancement and YOLOv8

Abstract

The advancement of automated approaches for detecting Pacific white shrimps in underwater environments is crucial for sustainable aquaculture practices. This study focuses on developing a novel methodology to enhance underwater shrimp images and detect shrimps using the YOLOv8 architecture. The specific objectives include developing a contrast-enhancement system to acquire clear underwater shrimp images and implementing the YOLOv8 model for shrimp detection. YOLOv8’s superior accuracy, speed, and memory efficiency allow non-intrusive detection of bottom-dwelling shrimps using computer vision and machine learning without disrupting their natural habitat. However, underwater images suffer from poor contrast and limited visibility due to light attenuation and scattering, making object detection and recognition challenging. Hence, a fusion-based underwater image contrast enhancement algorithm is employed to improve the shrimp detection accuracy. The enhanced images obtained through the contrast-enhancement system are utilised for length and weight estimation of the shrimps using the YOLOv8 model. The results of the study demonstrate promising outcomes, with precision of 71.7 %, recall of 86.8 %, and mAP 50 scores of 87.8 % in shrimp detection. This comprehensive approach not only enhances the accuracy of shrimp detection in farm ponds but also contributes to the overall health monitoring and management of shrimp populations. The methodology employed in this study combines image enhancement techniques with state-of-the-art deep learning models to improve the efficiency and precision of shrimp detection in underwater environments.