An adaptative classifier for the recognition of targets in CCD/CBERS images

• Main Authors: Antonio Roberto Formaggio, Yosio Edemir Shimabukuro, José Demísio Simões da Silva, Cléber Rubert, Viviane Todt
• Format: Article
• Language: English
• Published: Universidade Federal de Uberlândia 2006-12-01
• Series: Revista Brasileira de Cartografia
• Subjects: Artificial Neural Nets; CCD sensor; CBERS satellite; Araucariaangustifolia.
• Online Access: http://www.rbc.ufrj.br/_pdf_58_2006/58_03_9.pdf

This work presents an integrated neural classifier aiming to increase the accuracy in the recognition of different features inside CCD/CBERS images. Among these features there are native capons and areas of reforestation of Araucaria angustifolia located in the interior and in the proximities of the Sao Francisco de Paula National Forest. This forest is considered the older conservation unit from Rio Grande do Sul state, Brazil . The considered neural classifiers are said integrated because they are constituted of three models of neural nets grouped in two distinct approaches of integration: a) the two-third approach, and b) the credibility criterion approach. The used neural classifiers were: Multi-layer Perceptron, Learning Vector Quantization, and Radial Basis Function. The developed experiments showed that the integrated neural model using the credibility criterion approach contributes for increasing the accuracy in the identification of features, revealing promising for other applications, such as the monitoring in real time of the terrestrial surface.