A Modular MLP-Based Deep Feature Classification Framework for Wheat Cultivar Leaf Identification

Wheat cultivar identification is vital for optimizing yield and grain quality, as these are significantly influenced by the specific cultivar grown. Conventional classification approaches have predominantly relied on post-harvest seed characteristics, overlooking early growth stages where proactive decisions can have a substantial impact on crop performance. Early and reliable identification of cultivars empowers farmers with actionable insights, enhances food security, and supports precision-oriented crop management. To address this gap, the present study proposes an automated and modular deep learning framework for pre-harvest wheat cultivar identification using leaf images. A manually curated dataset of pre-harvest leaf images of ten prominent wheat cultivars was developed. Deep features were extracted using a pre-trained ResNet-\(\:18\) model, yielding \(\:512\)-dimensional representations for each image. These features were classified using a diverse set of multilayer perceptron (MLP) architecture variants. Among these, the residual-inspired MLP achieved the highest classification accuracy of \(\:99.16\text{\%}\), demonstrating effective discrimination of cultivars based solely on foliar traits. This framework offers an early, cost-effective, and scalable solution for cultivar recognition, with significant implications for precision agriculture, crop monitoring, and sustainable farm management.