Accurate classification of pumpkin seeds is critical in agriculture, impacting market value, quality control, and breeding decisions. However, traditional manual methods often fall short due to subtle morphological similarities among seed types. This study introduces a novel hybrid framework that leverages one-dimensional Convolutional Neural Networks (1DCNN) applied directly to handcrafted morphological features to improve classification performance. By learning hierarchical and discriminative representations from these pre-extracted features, the proposed 1DCNN effectively capture complex patterns overlooked by conventional approaches. We investigate two model variants: a deeper architecture with five convolutional layers and a lighter version with one fewer layer. Experimental results on a public pumpkin seed dataset show that our best-performing model achieves an accuracy of 90.52%, outperforming the previous benchmark of 88.64%. These findings highlight the potential of deep 1DCNN to enhance feature abstraction from tabular morphological data, offering a promising direction for automated and reliable agricultural product classification.