Enhancing YOLOv7 for underwater image object detection: a research study

Underwater images frequently experience color distortion and reduced visibility because of light absorption and scattering. To tackle these challenges, we introduce an improved algorithm for detecting objects in underwater images, ADM-YOLOv7. This algorithm uses the MLLE network to restore underwater images, improving their quality. Furthermore, the ResNet-ACmix module improves the model's capacity to gather comprehensive contextual information, boosting detection speed for smaller objects. Given the small and irregular nature of underwater targets, the DCN-E-ELAN module is designed to improve detection performance. The MPDIoU loss function further strengthens the model's ability to detect objects despite occlusions. Testing the improved ADM-YOLOv7 algorithm on the processed DUO dataset shows a significant increase in average precision from 79.82% to 84.00%, a 4.18% improvement, while reducing the average parameter count from 37.21M to 32.65M, a 12.25% decrease. Tr algorithms.