Anita Desiani, Deshinta Arrova Dewi, Dwiza Riana, Muhammad Arhami, Nuni Gofar, Fitri Salamah, Yunarni Pratiwi
Soil fertility is vital for agriculture because it affects plant growth and productivity. Automatic soil classification offers an effective solution for identifying soil types based on texture. Deep learning architectures are widely used to aid automatic classification. AlexNet is a deep learning architecture used for image classification. AlexNet can extract complex features but demands substantial memory and computational resources. AlexNet contains many neurons and parameters, which can increase the risk of overfitting. This study proposes BAXeNet, a novel architecture that integrates AlexNet with depthwise separable convolutions from XceptionNet and incorporates Batch Normalization to reduce overfitting and computational complexity. BAXeNet classifies soil into five texture based categories: Black Soil, Cinder Soil, Laterite Soil, Peat Soil, and Yellow Soil. Experimental results demonstrate that BAXeNet outperforms other models, achieving balanced average accuracy, precision, recall, Fl score, and ROC values ranging from 98% to 99%, with per class precision, recall, and Fl score between 90% and 99% BAXeNet balances model complexity and performance with 21.3 million parameters, delivering superior accuracy and the fastest training time (28 seconds), significantly outperforming other methods that require between 64 and 102 seconds. Overall, the evaluation confirms BAXeNet's robustness in texture based soil classification, effective model complexity, and computational efficiency. However, the recall for Peat Soil is relatively lower at 90%, highlighting the need for future work to improve classification performance for this class and achieve more consistent results across all categories. © (2026), (International Association of Engineers). All Rights Reserved.
Mathematics Department and Natural Science Faculty, Universitas Sriwijaya, Indralaya, 30662, Indonesia; Faculty of Data Science and Information Technology, INTI International University, Putra Nilai, 71800, Malaysia; Faculty of Information Technology, Universitas Nusa Mandiri, Jakarta, 13620, Indonesia; Faculty of Information Technology, Politeknik Negeri Lhokseumawe, Indonesia; Faculty of Agriculture, Universitas Sriwijaya, Indralaya, 30662, Indonesia; Computer Science Department, Universitas Sriwijaya, Palembang, 30139, Indonesia; Mathematics Department, Mathematics and Natural Science Faculty, Universitas Sriwijaya, Indralaya, 30662, Indonesia
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock