Objects recognition in aerial photographs with neural networks

The paper presents a part of a work concerning cadaster updating from aerial color photographs. The aim is to transmit to a database symbolic information about objects detected in the images. We use neural networks to give each object a label. We proceed in two major steps. In the first step, we send basic information about the three bands green, red and blue to specialized networks. Each of those networks is devoted to recognizing one kind of object. In the second step, we process a method to break ties, if any, in view to give each object one label. That procedure is tested on a sample including 335 objects