Hyperspectral Band Selection Considering Sensor Readout Speed and Classification Accuracy

Le résumé fourni par la source

We present a fast method to automatically select relevant hyperspectral bands for a specific classification task by finding the best compromise between sensor readout speed and classification accuracy considering sensor characteristics. To find that compromise, we have to consider the relation between sensor area and hyperspectral band. The method finds the most relevant spectral bands using feature importance of random forest together with dynamic programming for the optimization. At classification accuracies comparable to that of working with all bands, the band selections found with the proposed method allow for higher sensor readout speed and hence production line speed increases of 4 or more.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.