Underwater Object Detection using EfficientDet

Underwater object detection has numerous applications, including marine research, habitat inspection, mineral identification, and more. Taking high-quality photographs or videos underwater can be difficult because water absorbs and scatters light, which can make images appear unclear or distorted. As water depth increases, longer wavelengths of light are absorbed, resulting in predominantly blue-green images, while reds are absorbed due to their higher wavelengths. This leads to image degradation, including low contrast, color distortion, and poor visibility, making it challenging to detect objects in underwater images. Therefore, enhancing underwater images is crucial to improve their quality for various applications while retaining the valuable information they contain. However, low light conditions can negatively affect detection performance, and the image’s light composition significantly impacts subject detection. To enhance the detection outcomes, it is possible to utilize low-light image enhancement as a preliminary mechanism aimed at ameliorating the quality of the image Nevertheless, current enhancement methods may not work well in all situations due to the complexity of the wavelength environment. To tackle this issue, our solution involves a twostage process comprising of an enhancement phase followed by a detection phase. In the enhancement phase, we use multiple enhancement methods to automatically enhance low-light images and generate corresponding weights that are better suited for detection.