An Improved YOLOV5 Based on Triplet Attention and Prediction Head Optimization for Marine Organism Detection on Underwater Mobile Platforms

Abstract

Machine vision-based automatic detection of marine organisms is a fundamental task for the effective analysis of production and habitat changes in marine ranches. However, challenges of underwater imaging, such as blurring, image degradation, scale variation of marine organisms, and background complexity, have limited the performance of image recognition. To overcome these issues, underwater object detection is implemented by an improved YOLOV5 with an attention mechanism and multiple-scale detection strategies for detecting four types of common marine organisms in the natural scene. An image enhancement module is employed to improve the image quality and extend the observation range. Subsequently, a triplet attention mechanism is introduced to the YOLOV5 model to improve the feature extraction ability. Moreover, the structure of the prediction head of YOLOV5 is optimized to capture small-sized objects. Ablation studies are conducted to analyze and validate the effective performance of each module. Moreover, performance evaluation results demonstrate that our proposed marine organism detection model is superior to the state-of-the-art models in both accuracy and speed. Furthermore, the proposed model is deployed on an embedded device and its processing time is less than 1 s. These results show that the proposed model has the potential for real-time observation by mobile platforms or undersea equipment.

High-Precision Carton Detection Based on Adaptive Image Augmentation for Unmanned Cargo Handling Tasks

“…To overcome the problem of training set insufficiency, a Global Mask R-CNN detection algorithm based on a small training set was also presented by precisely composing the target feature region and saving the target semantic information in the deep learning backbone, and the precision could reach 66.45% [ 21 ]. For the one-stage series, the YOLOs are progressively proposed to improve the network structure, such as YOLO9000 [ 22 ], YOLOv3 [ 15 ] and YOLOv5 [ 23 ]. In one YOLOv3-based ship detection case, the detection precision could reach 55.3% [ 24 ].…”

“…By combining the CenterNet and YOLOv3 and introducing the spatial shuffle-group enhance (SSE) attention module, more advanced semantic features were integrated, avoiding the problem of detection omissions, and the precision was further improved to 90.6% [ 4 ]. On this basis, an extra detection head was added to the YOLOv5 model to improve the multi-scale detection and small target, experiencing an 11.6% rise [ 23 ]. In view of the better performance of YOLO series, this paper used YOLOv5 as the baseline to demonstrate the effectiveness of the proposed method.…”

Underwater Multi-Target Recognition Method Based on Improved YOLOv5

“…Yu et al [6] proposed an improved YOLOv4tiny detection algorithm, and the average detection accuracy of sea cucumber, sea urchin and other organisms about to 80%. Li et al [7] adopted the improved YOLOv5 model to detect sea urchins and other precious seafood, and the detection accuracy reached over 83.1%. The aforementioned studies suggest that the YOLO series especially YOLOv5 model exhibits potential detection accuracy and a more compact model structure in the realm of underwater biological detection, and thereby demonstrates enhanced suitability for target detection within complex underwater environments.…”

In Situ Sea Cucumber Detection across Multiple Underwater Scenes Based on Convolutional Neural Networks and Image Enhancements

“…A multi-scale feature fusion strategy, shortcut feature pyramid network (S-FPN) was proposed to improve the detection accuracy of small targets by introducing various shortcut connections between feature pyramids. Giving an additional detection head to the YOLOv5 model could also enhance the multiple scale detection and improve the detection accuracy of small targets [ 17 ]. Attention mechanisms were effective in learning object features and increasing the receptive field could improve the detection accuracy and robustness in recognizing underwater small targets [ 18 ].…”