How marine radars and firefly-inspired algorithms detect oil spills?

Oil spills in the sea threaten ecosystems, economies and human health. To fight them, researchers have created a new method that mixes marine radars, computer‑vision tricks and a firefly‑inspired algorithm.

First, radar images are cleaned: three steps sharpen the oil film’s edges, making it stand out. Then two visual “fingerprints” are extracted: Local Binary Patterns (LBP) that capture texture, and Histogram of Oriented Gradient (HOG) that describes edge directions. After a light median filter, the LBP and HOG data are merged into a rich feature map.

Next, a clustering algorithm called K‑Means separates suspicious zones from the rest of the image. The final step is classification: here the Improved Firefly Optimization (IFA), an upgraded firefly algorithm, decides whether a region is truly oil or just a radar artifact.

Compared with the classic FA, IFA offers higher precision in oil‑film segmentation, enabling faster and more reliable detection. This approach paves a new path for marine monitoring, allowing quick interventions and less environmental damage.

The proposed method follows a multi‑step processing chain. Using Local Binary Patterns (LBP) encodes the local texture of the radar image, exploiting intensity differences between neighboring pixels to highlight variations typical of an oil film, which tends to appear more homogeneous than the seabed. After extraction, mean filtering reduces high‑frequency noise while preserving relevant structures.

Histogram of Oriented Gradient (HOG) is then computed on the filtered LBP map. HOG divides the image into cells and calculates intensity gradients, building a histogram of dominant directions. This helps to differentiate the irregular edges of oil patches from other radar anomalies such as waves or floating objects.

K‑Means clustering operates in the combined feature space (LBP + HOG), minimizing intra‑cluster variance and maximizing inter‑cluster separation. The number of clusters is typically set to two (oil vs. non‑oil) but can be adapted for more complex scenarios.

Improved Firefly Optimization (IFA) introduces two enhancements over classic FA: (i) a dynamic brightness adaptation based on local feature density, and (ii) a mutation strategy that prevents trapping in local minima. Virtual fireflies (candidate solutions) move in the segmentation‑threshold space, attracted to brighter solutions (higher accuracy metrics). IFA converges faster and more robustly, especially under radar noise and low‑visibility conditions.

Experimental validation on real marine radar datasets shows segmentation accuracy exceeding 92%, improving by about 5‑7 percentage points over traditional FA and methods relying solely on LBP or HOG. These results suggest that integrating texture‑orientation features with evolutionary optimization could become a standard for marine environmental monitoring systems.