THE EFFECT OF NEURAL-NETWORK STRUCTURE ON A MULTISPECTRAL LAND-USE LAND-COVER CLASSIFICATION/

While neural networks are now an accepted alternative to statistical m ultispectral classification techniques for remote sensing image classi fication, the network approach presents both unique challenges and abi lities. The size of the hidden layer must be determined by trial and e rror, and the random initial weight settings result in different paths for the training procedure, making the network a non-deterministic cl assifier. For the sample classification presented here, it was found t hat there was a range of optimal hidden layer sizes below which the ac curacy decreased and above which the training time increased. However, it was also found that, for a fairly wide range, the hidden layer siz e made little difference to the final classification accuracy. Initial weight randomization was as much of a factor as hidden layer size. Us ing 3 by 3 windows of data in each band was found, despite increased t raining time per iteration, to achieve similar accuracy with less over all training time, although with less consistency.