Transparent Deep Neural Networks for High-Fidelity Species Identification in Microalgae
Türk Tarım - Gıda Bilim ve Teknoloji dergisi, cilt.14, sa.2, ss.533-543, 2026 (TRDizin)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 14 Sayı: 2
- Basım Tarihi: 2026
- Doi Numarası: 10.24925/turjaf.v14i2.533-543.8731
- Dergi Adı: Türk Tarım - Gıda Bilim ve Teknoloji dergisi
- Derginin Tarandığı İndeksler: CAB Abstracts, Central & Eastern European Academic Source (CEEAS), Directory of Open Access Journals, TR DİZİN (ULAKBİM)
- Sayfa Sayıları: ss.533-543
- Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
- Atatürk Üniversitesi Adresli: Evet
Özet
Microalgae are morphologically diverse and highly plastic, making species-level identification challenging; thus, reliable automated classification is crucial for ecological and bioresource applications. This study aims to develop a transparent and biologically interpretable deep learning framework for microalgae species recognition using explainable artificial intelligence (XAI). Microscope images of three phylogenetically distinct taxa Cryptomonas ovata (Cryptista), Ceratium hirundinella (Pyrrophyta), and Tetradesmus dimorphus (Chlorophyta) were analyzed using eight convolutional neural networks from the EfficientNet family (B0–B7). All models were trained via transfer learning using ImageNet-pretrained weights and standardized preprocessing and augmentation pipelines. Classification performance was evaluated on independent test datasets using accuracy and loss metrics, and model interpretability was assessed through saliency maps and Gradient-weighted Class Activation Mapping (Grad-CAM). EfficientNet-B2 and B3 achieved the strongest performance, reaching 99.45% validation accuracy and 98.72% test accuracy, while other EfficientNet variants also demonstrated consistently high predictive reliability. XAI visualizations revealed that both Grad-CAM and saliency maps emphasized biologically meaningful cellular structures such as cell walls, lorica boundaries, and thecal extensions indicating that the models relied on taxonomically relevant morphological cues rather than background artifacts. This interpretability confirms that the network’s decisions are grounded in recognizable diagnostic traits. These findings show that deep learning with XAI can deliver accurate and interpretable species-level identification of microalgae, supporting automated taxonomy and broader aquatic research applications.