Then, image objects corresponding to potential ship targets are created by means of a region�Cgrowing clearly algorithm.During the feature extraction stage, image objects are characterized by spectral, shape and textural features. The feature extraction selleck chemical Belinostat Inhibitors,Modulators,Libraries step is concerned with finding transformations to map features to a lower dimensional space for enhanced class separability and optimized performance. Because of their robustness, GAs are considered a suitable tool to address the optimization problem [9]. The GA�Cdriven selection procedure provides a vector of feature values corresponding to a series of feature combinations that is passed to the subsequent classification stage.In the third and Inhibitors,Modulators,Libraries last step, artificial neural networks are used for the classification of image objects.
A neural network architecture is created according to the optimal feature combinations and optimal number of hidden nodes. Figure Inhibitors,Modulators,Libraries 1 shows an overview of Inhibitors,Modulators,Libraries the approach which Inhibitors,Modulators,Libraries consists of two phases: a learning phase and an operational phase.Figure 1.Flow chart of the proposed ship detection algorithm implemented in two phases: a learning phase and an operational phase.In the learning phase, GAs are used to train a feed�Cforward neural network based on reference samples. An objective function is used to calculate fitness that is equal to the inverse of classification error rate. In the operational phase, the best low dimensional neural network architecture is selected as a classifier in the three�Cstep ship detection and classification algorithm.
In what follows, the three main stages of the algorithm are described in detail.
2.2. Segmentation (Pre�Cdetection of ship patterns)Pre�Cscreening of possible ship patterns Inhibitors,Modulators,Libraries is based on the contrast between Inhibitors,Modulators,Libraries sea (noise�Clike background) Inhibitors,Modulators,Libraries and target (a cluster of bright pixels). The contrast depends on the sea conditions, the ship’s detailed shape, and its position relative to the satellite beam. The proposed algorithm applies a 100 Entinostat x 100 pixel moving window adaptive threshold to the image pixel values (Xi,j) to discriminate bright pixels [10]. The threshold used for the detection of intensity peaks is based on the mean (��oc) and the standard Anacetrapib deviation (��oc) of the sea background in the moving window.
Xi,j?��oc��oc��Threshold(1)Noise resulting from image thresholding is removed using a morphological opening operation with a 2 �� 2 pixels structural element.
Indeed, isolated pixels cannot belong to a ship object, which is usually characterized by a cluster of several bright pixels.The resulting thresholded image CC-5013 is then segmented into coherent image objects by means of the region�Cgrowing segmentation. click here Shrimp boats have a characteristic shape, usually consisting of two regions of high intensity related by a region of lower intensity as shown in figure 2. The region�Cgrowing operator allows the grouping of the regions of a ship that may be detected separately during the thresholding operation.Figure 2.