Feedback Convolutional Network for Intelligent Data Fusion Based on Near-infrared Collaborative IoT Technology

Near-infrared (NIR) spectral data has response information to target composition, sparsely implied in spectral frequency sequence. An internet of things (IoT) framework constructed with NIR calibration platform needs advanced algorithms to realize intelligent analysis. In this work, a feedback convolutional neural network (CNN) architecture is designed to extract spectral features. The architecture includes three repeated segments for multiple extraction of NIR spectral features. Error-feedback iteration is proposed to optimize the convolution filters of each segment. The multi-segment features are fused to train the calibration models. The proposed feedback CNN architecture for information fusion is applied to NIR analysis of selenium in paddy rice. Experimental results show that the fusion of multi-segment features is prospective to enhance the ability of NIR calibration. The feedback convolutional network for information fusion is expected to be applied in the NIR collaborative IoT framework, to ensure high performance in intelligent analysis of the IoT.