Computer Vision-Based Maize Seed Classification Using Multi-Head Self-attention Gated-Dilated Convolutional Neural Network and Res2NeT

Maize or corn seeds are the reproductive organs of Zea mays, the corn plant. Maize is a cereal grain that is widely grown because its edible seeds are a staple diet in several parts of the world. In this paper, a Computer Vision-based Maize Seed Classification using a Multi-Head Self-Attention Gated-Dilated Convolutional Neural Network and Res2NeT (MSE-CV-MSGCNNNET) is proposed. Here, the input images are gathered from the Maize seeds dataset. The input image is pre-processed using adaptive multi-scale improved differential filter (AMSIDF) and removing the maize shadows and cleaning background noise present in the image; then the pre-processed image is provided to segmentation using Multi-Kernel Collaborative Fuzzy Clustering (MKCFC) for selecting the appropriate threshold, which segments the binary image into multiple parts. After segmentation, the images are fed to the feature extraction stage utilizing Ternary Pattern and Discrete Wavelet Transform (TPDWT) to extract features such as dimensional and shape features. The Optimized Multi-Head Self-Attention Gated-Dilated Convolution Neural Network with Res2NeT (MSGCNNNET), which is used for accurate maize seed classification of LiaoGe varieties, XinNuo varieties, and KouXian varieties. Finally, the Intensified Sand Cat Swarm Optimization (ISCSO) algorithm is employed to optimize the MSGCNN-Net, which classifies the maize seeds accurately. The MSE-CV-MSGCNNNET is examined under some metrics: accuracy, precision, specificity, recall, f1-score, computation time and ROC. The proposed MSE-CV-MSGCNNNET approach attains 98.63% higher accuracy and 98.55% higher precision when compared to existing models.