Hybrid Deep-Machine Learning for Butterfly Species Image Classification

Authors

• Maha A. Rajab, Department of Biology Science, College of Education for Pure Sciences/Ibn AL-Haitham, University of Baghdad, Baghdad, IraqFollow
• Wisal Hashim Abdulsalam, Department of Computer, College of Education for Pure Science (Ibn Al-Haitham), University of Baghdad, Baghdad, IraqFollow
• Firas A. Abdullatif, Department of Computer, College of Education for Pure Science (Ibn Al-Haitham), University of Baghdad, Baghdad, IraqFollow
• Tole Sutikno, Department of Electrical Engineering, Faculty of Industrial Technology, Universitas Ahmad Dahlan, Yogyakarta, IndonesiaFollow

Abstract

Classifying butterfly species is crucial in biodiversity studies and environmental monitoring. However, manual classification is often a laborious process that requires specialized expertise and is prone to error, especially when species have similar visual characteristics. To address these drawbacks, this paper presents a hybrid approach that combines machine learning with deep learning for feature extraction. To enhance the visibility of important features, preprocessing techniques such as background removal and binarization are applied to butterfly images. Feature extraction was performed using the SqueezeNet convolutional neural network, pretrained on the ImageNet dataset. By discarding the final classification layer, the network produced discriminative feature vectors that effectively captured the visual attributes of each butterfly. These feature vectors were then used to train a range of machine learning classifiers, including Support Vector Machines (SVM) with different kernels, k-Nearest Neighbors (KNN), Neural Networks (NN), and Stochastic Gradient Descent (SGD). The dataset utilized in this study is designed for identifying butterfly and moth species. It includes 100 distinct category labels. According to these findings, the SVM with a polynomial kernel achieved the best classification accuracy (up to 95%), while the NN followed closely. However, KNN and SGD had somewhat lower accuracy, and external testing of the framework on unknown images was found to be 83.5% on images taken outside the dataset. Hence, the effectiveness of deep learning in feature extraction is demonstrated when combined with machine learning classifiers.

Keywords

Butterfly classification, Deep learning, Machine learning, Neural networks, SVM, SqueezeNet

Subject Area

Computer Science

Article Type

Article

First Page

3364

Last Page

3376

Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 International License.

How to Cite this Article

Rajab, Maha A.; Abdulsalam, Wisal Hashim; Abdullatif, Firas A.; and Sutikno, Tole (2026) "Hybrid Deep-Machine Learning for Butterfly Species Image Classification," Baghdad Science Journal: Vol. 23: Iss. 9, Article 22.
DOI: https://doi.org/10.21123/2411-7986.5416