Integration of Computer Vision with Deep Feature Representation Learning for Robust Wheat Seed Variety Identification and Purity Analysis

Abstract

Wheat is the primary staple grain in temperate climates and is increasingly sought after in countries experiencing growth in their cities and industries. It is a primary food source that provides essential energy, protein, and other nutrients, and its rising demand is driven by factors like population growth and changing dietary habits. Identifying and classifying seed varieties is even conducted physically over direct visual assessment, which is time-consuming, labour-intensive, and prone to error. Thus, it is vital to develop a sophisticated, automated method that is economical and rapid to improve agricultural manufacture and grain integrity. To attain this goal, computer vision and artificial intelligence techniques have attained higher precision in detecting and classifying the complex seed features. Since the advent of convolutional neural networks (CNNs), several studies have focused on applying deep learning methods in the domain of agricultural. Deep learning is an advanced model for image processing and data analysis. Therefore, this study introduces an Automated Wheat Seed Variety Classification and Purity Assessment using the Optimal Deep Learning (AWSVCPA-ODL) technique. The purpose of the AWSVCPA-ODL technique is to accurately analyse wheat seed images using computer vision and deep learning, enabling robust varietal identification and reliable purity assessment. At the initial stage, the proposed AWSVCPA-ODL technique follows two pre-processing steps: Gaussian filtering-based noise elimination and contrast enhancement to improve the quality of input images for further analysis. For feature extraction, the AWSVCPA-ODL technique employs a feature fusion process that combines EfficientNet, SqueezeNet, and a Capsule network. A bidirectional gated recurrent unit with self-attention is applied for the detection and classification of wheat variety and its purity. To further enhance the performance of classification model, the Adabelief hyperparameter optimizer is applied. To ensure the better performance of proposed AWSVCPA-ODL method, a wide range of simulations were performed on the benchmark database. The results indicated that the proposed technique outperformed existing models.