Comparative analysis of deep learning architectures for multi-label classification of wolffia globosa conditions in mobile smart farming

Name: Phae, Kyar

Organization : King Mongkut’s University of Technology Thonburi. School of Information Technology

Name: Pongkhun Poonsupsopon

Organization : King Mongkut’s University of Technology Thonburi. School of Information Technology

Name: Pavadol Dhechasidphaisan

Organization : King Mongkut’s University of Technology Thonburi. School of Information Technology

Name: Pornchai Mongkolnam

Organization : King Mongkut’s University of Technology Thonburi. School of Information Technology

Name: Watanyoo Suksa-ngiam

Organization : King Mongkut’s University of Technology Thonburi. School of Information Technology

Email : watanyoo.suks@kmutt.ac.th

keyword: Convolutional neural networks

LCSH: Duckweeds -- Disease and pest resistance

; Smart farming

LCSH: Deep learning (Machine learning)

LCSH: Mobile apps

Abstract: Wolffia globosa (duckweed) is emerging as a high-value sustainable protein source, yet its commercial viability is contingent on meeting strict food safety standards. While the crop is robust, ensuring high-quality, contaminant-free biomass is difficult for small-scale farmers who lack the expertise to visually diagnose subtle pond health issues such as nutrient imbalances or bacterial contamination. To address this, we propose a mobile-based automated diagnosis system. This study evaluates the feasibility of using deep learning models for multi-label classification of pond conditions for Wolffia globosa in a resource-constrained mobile environment. We curated a new dataset of pond images representing six distinct conditions, collected in a controlled experimental setup, and augmented to address class imbalances. We benchmarked seven Convolutional Neural Network (CNN) architectures, including EfficientNet (B0, B2, V2B0), ResNet50, MobileNetV3Large, VGG16, and AlexNet, using K-Fold cross-validation. Results indicate that EfficientNetB2 achieved the highest performance, with a K-Fold average accuracy of 0.7850 and a macro F1-score of 0.95. While standardized training stabilized deeper architectures like ResNet50, EfficientNetB2 remained the optimal choice, significantly outperforming legacy, non-residual architectures such as VGG16 and AlexNet. These results indicate that modern lightweight CNNs are highly effective for rapid visual diagnosis of farmer-captured images, forming the foundational core of a smart farming application.

King Mongkut's University of Technology North Bangkok. Central Library

Address: BANGKOK

Email: library@kmutnb.ac.th

Created: 2026

Modified: 2026-08-04

Issued: 2026-08-04

บทความ/Article

application/pdf

BibliograpyCitation : In ECTI Association and Burapha University. Faculty of Engineering. 2026 23rd International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology (ECTI-CON 2026) (pp.92-97). Piscataway, NJ : IEEE

eng

©copyrights King Mongkut's University of Technology North Bangkok

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	ECTI-CON 2026pp.92-97.pdf	3.18 MB

ใช้เวลา

0.036776 วินาที