Biogeography and geo-sciences based land cover feature extraction

• Major assumption-entropy is the driving force similar to the convection forces.
• Outperforms the hybrid ACO2/PSO/BBO classifier.
• Classifies homogeneous regions more efficiently than others developed till date.
• Proves to be the best known classifier developed till date.

This work proposes a biogeography and geo-sciences based soft computing technique which is an extension of original biogeography based feature extraction algorithm using the concept of entropy inspired from the geo-sciences phenomenon of mantle convection and dynamics of the earth. This algorithm uses surface entropy in the relevant band of multi-spectral images as the basis of calculating the habitat suitability index which in turn forms the basis of identifying different terrain features in the satellite image. The proposed work has been primarily developed for the purpose of finding the applications of geo-sciences in developing computationally intelligent models. This may lead to another concept of process randomization, generation of virtual scenarios, etc. which are important ingredients in battlefield assessment. The proposed feature extractor algorithm has been applied on the datasets of Alwar region in Rajasthan and Patalganga area in Shivalik ranges. The results indicate that our proposed geo-sciences based classifier is highly efficient in extracting land cover features. Further when integrated with hybrid bio-inspired intelligent classifier proposed in our previous work, it improves its classification efficiency and outperforms the earlier probabilistic classifiers, recent soft computing classifiers such as membrane computing, hybrid FPAB/BBO, extended non-linear BBO, etc. and the very recent hybrid ACO2/PSO/BBO classifier proposed by us [16] and [21]. Our results conclude that the classifier based on our proposed model is the best known classifier developed till date. The proposed model is flexible and can adapt itself to suit to a large number of classification problems including mixed pixel resolution, face recognition, pattern recognition, etc. whereby entropy can be simply calculated in any other band or according to its standard definition and hence feature extraction can be made.

There are two underlying concepts which form the base of the geo-sciences phenomenon, the first being the dynamics of the movement of earth's plates and the second is the crust formation [9], [22] and [35]. We consider that it is the entropy or the degree of disorder of the earth's mantle convection which is the driving force of the mechanism of plate dynamics. Plate dynamics is considered as analogous to the clustering of similar pixels which is accomplished through texture analysis using rough sets in our proposed algorithm and resulting crust formation is considered as analogous to the presence of mixed pixels in an input plate. Hence, the underlying concepts of geo-sciences can be adapted to integrate with the original BBO technique for feature extraction in the satellite image. This paper is an attempt to modify the original biogeography based feature extraction technique using an analogy between biogeography and geo-sciences phenomenon of earth's dynamics as shown in the figure below and some of the ideas of which were first presented in our paper [18].Figure optionsDownload as PowerPoint slide