Automated End-to-End Deep Learning Framework for Complex Multiclass Brassica Seed Classification

Abstract

Agricultural research has accelerated in recent years, but farmers often lack the time and resources to conduct on-farm experiments, as most of their efforts are devoted to crop production. Seed classification provides essential insights for seed quality control, impurity detection, and yield estimation. Early identification of seed types is critical to reduce costs, minimize risks of poor field emergence, and support efficient crop management. Traditional classification methods rely heavily on manual feature extraction and expert input, which limits scalability and accuracy when dealing with highly similar seed types. To address this challenge, we propose an automated end-to-end deep learning framework for complex multiclass Brassica seed classification. Our framework integrates preprocessing, feature learning, and classification into a unified pipeline, eliminating the need for handcrafted features. Using a newly collected dataset of ten Brassica seed classes characterized by high texture similarity, we develop and evaluate a convolutional neural network optimized through architectural design and hyperparameter tuning. Experimental results demonstrate that the proposed framework achieves a classification accuracy of 93%, outperforming several state-of-the-art pretrained models. These findings highlight the potential of automated end-to-end deep learning models to enhance precision agriculture, providing robust and scalable solutions for seed quality monitoring and agricultural productivity.