DualCMNet: A Lightweight Dual-branch Network for Maize Variety Identification based on Multi-modal Feature Fusion

Xinhua Bi¹Hao Xie¹Ziyi Song²Jinge Li¹Chang Liu¹Xiaozhu Zhou¹Helong Yu¹Chunguang Bi^1*Ming Zhao^3*

• ¹College of Information Technology, Jilin Agriculture University, Changchun, China
• ²School of Electronic Information Engineering, Changchun University of Science and Technology, Changchun, Jilin Province, China
• ³Jilin Zhongnong Sunshine Data Company Limited, Changchun, Hebei Province, China

The accurate identification of maize varieties is of great significance to modern agricultural management and breeding program. However, traditional maize seed classification methods mainly rely on single modal data, which limits the accuracy and robustness of classification. In addition, existing multimodal methods face high computational complexity, making it difficult to balance accuracy and efficiency. Based on multi-modal data from 11 maize varieties, this paper presents DualCMNet, a novel dual-branch deep learning framework that achieved accurate maize seed variety classification by utilizing a one-dimensional convolutional neural network (1D-CNN) for hyperspectral data processing and a MobileNetV3 network for spatial feature extraction from images. Furthermore, to enhance feature expression capability and reduce computational overhead, the framework introduces three key improvements: the HShuffleBlock feature transformation module implements feature dimension alignment and information interaction; the Channel and Spatial Attention Mechanism (CBAM) enhances the expression of key features; and the lightweight gated fusion module dynamically adjusts feature weights through a single gate value to improve fusion effects. During the training process, this study utilizes pre-trained 1D-CNN and MobileNetV3 models for network initialization and implements a staged training strategy, first optimizing non-pre-trained layers, then unfreezing pre-trained layers and applying differentiated learning rates for fine-tuning. Through 5-fold cross-validation evaluation, the method achieved a classification accuracy of 98.75% on the validation set, significantly outperforming single-modal methods, while the total model parameters are only 2.53M, achieving low computational overhead while ensuring high accuracy. This lightweight design enables the model to be deployed in edge computing devices, allowing for real-time identification in the field, thus meeting the practical application requirements in agricultural Internet of Things and smart agriculture scenarios. This study not only provides an accurate and efficient solution for maize seed variety identification but also establishes a universal framework that can be extended to variety classification tasks of other crops.