Neural network structure generation for the classification of remotely sensed data using simulated annealing | |
| Author | Karunasekera, H. N. D. |
| Call Number | AIT Thesis no. CS-92-4 |
| Subject(s) | Remote sensing--Data processing Image processing--Remote sensing |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering, School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Abstract | Artificial feedjorward neural networks are increasingly being used for the classification of remotely sensed data. However the back-propagation algorithm, which is so widely used does not provide any information about the structure of the network for a given classification problem. A network too small for a given problem will cause back-propagation not to converge to a solution while larger ones lead to poor generalization. Minimum sized networks are important for efficient classification and good generalization. The technique proposed in this work, simultaneously learns the structure and the weights of the network from the information present in training data/or a g.iven classification problem by minimizing a cost function which is proportional to the total error over the pattern set, to the number of nodes and connections in the network. Simulated Annealing is being employed as a technique to perform the optimization due to it's ability to optimize very complex optimization problems. |
| Year | 1992 |
| Type | Thesis |
| School | School of Engineering and Technology (SET) |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Computer Science (CS) |
| Chairperson(s) | Hosomura, T. |
| Examination Committee(s) | Huynh, Ngoc Phien ;Sadananda, Ramakoti |
| Scholarship Donor(s) | Government of Finland ; |
| Degree | Thesis (M.Eng.) - Asian Institute of Technology, 1992 |