Texture Features and Unsupervised Learning-Incorporated Rain-Contaminated Region Identification From X-Band Marine Radar Images

A novel method is proposed for identifying rain-contaminated regions in X-band marine radar images. Due to the difference of texture between rain-contaminated and rain-free echoes, a Gabor filter bank and discrete wavelet transform (DWT) are introduced to filter marine radar images and generate texture features. Feature vectors extracted from each pixel of the training samples are input into a clustering model, which is trained using unsupervised learning techniques such as k-means and a self-organizing map (SOM). After distinguishing between rain-free and rain-contaminated clusters, the proposed method is able to cluster pixels into rain-free and rain-contaminated types automatically. Images collected from a shipborne marine radar in a sea trial off the east coast of Canada under rain conditions are utilized to validate the proposed method. Identification results obtained from several clustering models with different combinations of cluster number, texture features, and clustering methods show that rain-contaminated pixels are effectively detected, with an overall identification accuracy of 89.1% for both k-means-based (k = 4) and 2 x 2-neuron SOM-based clustering models.