A Feature Preprocessing Framework of Remote Sensing Image for Marine Targets Recognition

The effective extraction of continuous features of marine remote sensing image is the key to the processing of marine target recognition. Since many of the existing data mining algorithms can only deal with discrete attributes, it is necessary to convert continuous features into discrete features to adapt to these intelligent algorithms. In addition, most of the current discretization algorithms do not consider the mutual exclusion of attributes as well as that of breakpoints within an attribute when selecting breakpoints, and cannot guarantee the indistinguishable relation of decision table. So, it is not suitable for dealing with remote sensing data with multiple features obviously. Aiming at these problems, a feature preprocessing framework of remote sensing image for marine targets recognition is proposed in this paper. In the frame design of the whole preprocessing algorithm, the equivalent relationship model of information entropy is introduced to perform a series of comparison operations and loop controls, so as to obtain the optimal discretization interval number. Finally, simulation analysis is conducted on the high-resolution remote sensing image data collected in the port areas of South China Sea. The experiment shows that the framework proposed in this paper achieves excellent results in terms of interval number, accuracy, running time, and effectively detects the vessels targets at sea. Therefore, the proposed framework can be very well applied to the discretization of remote sensing image features for marine targets recognition.