**6. Classification using features of oil slick**

In general, oil spill detection could be carried out either by manual process using trained personnel or by an automated system. As per literature review, existing systems uses either manual approach or semi-automated system. The numbers of SAR images to be analyzed are getting increased annually. If oil spill detection is carried out manually in a wide swath image with appropriate resolution, it is a very timeconsuming process. Thus, it is required to design an automated system for improving results. In certain cases, as outlined in [31], it is observed that manual detection is preferred.

The manual oil spill detection approach at KSAT is described in [61]. About the wind speed Automatic Information System (AIS) ship tracking of sea and their outer surface information, location of oil rigs, pipeline, national territory borders, and coastlines are available to the operator. Possible oil spills are assigned confidence levels based on their contrast to the surroundings, wind speed, identification of possible sources, natural slicks or low-wind areas nearby, and edge and shape characteristics of the slick. Various algorithms for oil spill detection based on single-polarization SAR images are reviewed in [62–64]. Several of the papers describe a methodology consisting of identifying dark spots followed by computing features describing the shape, contrast, surroundings, and homogeneity of the spots. The main challenge is not to segment out dark areas in the SAR image, but to identify a set of good features that can be used to discriminate between oil slicks and look-alikes, and then to use the features in a reliable classifier.

Some of the algorithms for oil spill detection based on SAR polarization images were reviewed from a single direction. Discernibly, the huge uniqueness spill and sea *Computational Techniques of Oil Spill Detection in Synthetic Aperture Radar Data: Review Cases DOI: http://dx.doi.org/10.5772/intechopen.108115*


#### **Table 5.**

*SAR image single-polarization features.*

surface territories are checking of extensive regions for oil detection. The dissimilarity decreases about the generality of wind levels. But the main complicating thing is the low wind dark area with very high dissimilarity common in low-wind conditions. Most important features are, if possible, spills from the selected region can be seen on SAR image. Polarimetric SAR image features can be extended to a certain region and availability of SAR-limited data.

The existing system feature extraction for the oil spill detection is given in [63]. Most commonly used features are shown in **Table 5** for single-polarization SAR oil-spill detection and **Table 3** features are used for polarimetric SAR oil spill detection [64]. The most of polarimetric SAR features are listed in **Table 6**. These are not used for automatic algorithms. But some automatic methods are to be used for detection of oil spill with polarimetric SAR data. The main comparison between the accuracy with single SAR polarimetric and polarimetric SAR data is not yet processed for large datasets.

The existing system experimental results are tried to get the rank features according to their importance. The objective was to select and use only those features characterized by strong discriminative capacity. It appears that combinations of features with high discriminative capacity were not giving satisfying results as combinations of features having lower discrimination capabilities. For example, if we have 10 cases, three of which can be discriminated by feature X, the question is how many of the remaining seven cases can be discriminated by another feature, for example, feature


**Table 6.**

*Polarimetric features used for oil spill detection.*

Y. If features X and Y contribute only in discriminating the same three cases, then the combination is not good and another feature (e.g., feature Z) has to be used.
