  ｦ   ｸ ｸ ｹ ｷ ｨ ｨ ｩ ｧ  c j q jk ik ik d d m 1 1 / 2 1 ｦ ｦ ｦ    ｸ ｸ ｹ ｷ ｨ ｨ ｩ ｧ    c i C k i k c i i i k J J 1 2 , 1 u c u AUTOMATIC DETECTION OF OCEANIC EDDIES IN SeaWiFS-DERIVED COLOR IMAGES USING NEURAL NETWORKS AND SHAPE ANALYSIS Samarth Patel, Ramprasad Balasubramanian, Avijit Gangopadhyay Department of Computer and Information Science School of Marine Science and Technology University of Massachusetts at Dartmouth, Dartmouth, MA r.bala@umassd.edu Avijit@umassd.edu Key remote sensing instruments like Advanced Sea-viewing Wide Field-of-view Sensor (SeaWiFS) aboard satellites, play a vital role in collecting observations that help in analyzing the properties of oceans around the globe. This research focuses on analysis and processing of high-resolution chlorophyll (ocean color) observations from SeaWiFS to automatically identify and segment ocean features. An oceanic eddy is a circular or whirling flow of water generally found along the edge of a prominent current. Mesoscale eddies are those vortices, whose diameters ranges from a few kilometers in the coastal ocean to a few hundred kilometers in the deeper open ocean. The objective of this work is to automatically detect and segment oceanographic eddies from chlorophyll color images. The focus of this research is on the Monterey Bay region off the California coast. The input to this system is the SeaWiFS color (Chlorophyll) data, along with the Neural Net (NN), trained on clustered Fuzzy C-means classification. The Fuzzy C-Means algorithm is a clustering method in which each piece of data could belong to two or more clusters. It is based on minimization of the following objective function: Fuzzy partitioning is carried out through an iterative optimization of the objective function shown above, with the update of membership u ij and the cluster centers c j by: where m is any real number greater than 1, uij is the degree of membership of xi in the cluster j, xi is the ith of d-dimensional measured data, cj is the d- dimension center of the cluster, and ||*|| is any norm expressing the similarity between any measured data and the center (we use the distance measure). The Fuzzy C-Means algorithm allows a point to be on the boundary [5]. Much time was spent on training and selecting an optimum neural network for the clustering of the chlorophyll data. The artificial neural network was trained in such a manner that it learned from all possible scenarios like cloudy images, cloud free images, data with maximum range of chlorophyll value. Combination of the data of various days was done such that each of the above scenarios was accounted for. The artificial neural net composed of one input layer, 2 hidden layers and one output layer. 30 nodes in each hidden layer were chosen for the simulation to obtain an optimal performance goal of 0.0001. The output of the NN, which is a labeled image, where the regions are combined based on proximity and