A Lightweight Hybrid Network for Agricultural Origin Identification Combined with Electronic Nose and Hyperspectral Systems

Abstract

Identifying the origin of agricultural products has an important practical significance for protecting consumers' rights and interests, maintaining the reputation of regional brands, and cracking down on unscrupulous merchants. Multimodal information fusion technology can comprehensively capture the characteristics of samples, offering an approach to solve the problem of identifying the origin of agricultural products. In this paper, a Dual-Modal Complementary Fusion Network (DMCF-Net) is proposed to precisely identify the origins of agricultural products. Using electronic nose and hyperspectral in order to fully explore the complementarity of gas and spectral information, we design a Cross-information fusion (Cif) module in DMCF-Net, which fuses gas and spectral information to obtain more comprehensive agricultural origin characteristics. Furthermore, to further improve the efficiency of feature utilization, we introduce the Self-Attention and Convolution mixed (ACmix) module after the Cif module to capture the local and global dependencies of features. The effectiveness of DMCF-Net is evaluated on two different datasets: 99.50% accuracy, 99.55% precision, and 99.34% recall on rice origin, and 99.72% accuracy, 99.75% precision, and 99.67% recall on peanut origin. The results prove that DMCF-Net is an efficient and lightweight recognition network, which provides an effective method for the identification of agricultural products' origin.

Keywords: Agricultural origin identification, Electronic nose, Hyperspectral, Data fusion, Convolutional neural network