3DeepM: An Ad Hoc Architecture Based on Deep Learning Methods for Multispectral Image Classification

Current predefined architectures for deep learning are computationally very heavy and use tens of millions parameters. Thus, computational costs may be prohibitive many experimental or technological setups. We developed an ad hoc architecture the classification multispectral images using techniques. The architecture, called 3DeepM, is composed 3D filter banks especially designed extraction spatial-spectral features in multichannel images. new has been tested on a sample 12210 seedless table grape varieties: Autumn Royal, Crimson Seedless, Itum4, Itum5 Itum9. 3DeepM was able to classify 100% obtained best overall results terms accuracy, number classes, parameters training time compared similar work. In addition, this paper presents flexible reconfigurable computer vision system acquisition range 400 nm 1000 nm. enabled creation first dataset consisting 37-channel (12 VIS + 25 IR) five varieties that have used validate architecture. Compared such as AlexNet, ResNet with high parameters, shows performance despite 130-fold fewer than which it compared. can multitude applications images, remote sensing medical diagnosis. small make ideal application online systems aboard autonomous robots unmanned vehicles.