Underwater Marine Debris Classifier

Overview

Constructed a specialized deep learning computer vision model to identify and locate submerged pollutants, nets, plastics, and debris from live ROV camera feeds in real time. Designed specifically to handle underwater optical challenges such as light attenuation, particulate backscatter, and color distortion.

Architecture & Features

• Underwater Image Preprocessing: CLAHE (Contrast Limited Adaptive Histogram Equalization) and chromatic channel balancing to compensate for red light absorption in ocean depths.
• YOLO Deep Learning Backbone: Custom fine-tuned weights for marine debris categories with high mean Average Precision (mAP).
• Edge Inference: Optimized ONNX Runtime export for deployment on low-power onboard companion computers inside underwater vehicles.

Source Code & Verification