The IRMA Community
Newsletters
Research IRM
Click a keyword to search titles using our InfoSci-OnDemand powered search:
|
Classification of Environmental Microorganisms for Monitoring Water Quality for Food Safety Using Deep Learning
|
|
Author(s): Anusha Tripathi (St. Xavier's College, Kolkata, India), Khushi Purohit (St. Xavier's College, Kolkata, India), Pragya Sinha (St. Xavier's College, Kolkata, India), Debabrata Datta (St. Xavier's College, Kolkata, India)and Anal Acharya (St. Xavier's College, Kolkata, India)
Copyright: 2026
Pages: 48
Source title:
Detection, Identification, and Monitoring of Food Contaminants
Source Author(s)/Editor(s): Matthew Chidozie Ogwu (Appalachian State University, USA)and Sylvester Chibueze Izah (Bayelsa Medical University, Nigeria)
DOI: 10.4018/979-8-3373-3982-5.ch008
Purchase
|
Abstract
Water contamination poses a huge risk in both food production and consumption. To monitor water quality, environmental microbes (EMs) can be used as they act as bioindicators. Automating the detection of EMs would be a highly time-saving and vital procedure, particularly since early monitoring of environments can prevent large-scale agricultural contamination. However, such automation remains underdeveloped for industrial use. Hence, a series of comparison experiments were conducted on various deep learning models using the open-source dataset EMDS-6. The pipeline includes segmentation followed by classification using different Convolutional Neural Networks (CNNs) to achieve a fully automated system. Given the limited dataset size, transfer learning, data augmentation, and k-fold cross-validation to mitigate overfitting and bias are employed. It was observed that deeper models like DenseNet121 achieve the highest accuracy (71.67%), whereas lightweight architectures such as MobileNetV2 strike a balance between accuracy and computational efficiency, making them well-suited for resource constrained environments. Finally, feature fusion is applied on the classifier models to assess potential improvements in accuracy which concatenates the feature representations from individual models. While fusion did not consistently outperform standalone models and often showed minimal or negative improvement, combining MobileNetV2 with other models improved performance in most cases. The fusion of Xception and MobileNetV2 yielded an accuracy improvement of 1.90%, while DenseNet121 and MobileNetV2 achieved the highest overall accuracy (71.19%) among all combinations.
Related Content
|
Rudrajit Paul, Meera Suresh.
© 2026.
40 pages.
|
|
Jayasree S. Kanathasan.
© 2026.
22 pages.
|
|
Udugalage Isuru Harsha Kumara, Welisarage Crista Ridmie Sumilaka Fernando, Gunavathy Selvarajh.
© 2026.
26 pages.
|
|
Poonam Maurya, Neelesh kumar Maurya, Nidhi Bhatt.
© 2026.
34 pages.
|
|
Jyoti Prakash, Urvashi Kumari.
© 2026.
24 pages.
|
|
Yassine Mouniane, Soumaya Choukri, Soukaina Elkadaoui, Issam El-Khadir, Ahmed Chriqui, Abdelaati Soufiani, Sara Razzak, Hiba Benchehida, Mariame Taibi, Rida Arabi, Fatima Kazbekovna Serebryanaya, Ilham Zenouaki, Mohamed El Bakkali, Mohamed Rektouti, Taoufiq Benali, Driss Hmouni.
© 2026.
30 pages.
|
|
Amira A. Gamal, Amal M. Hashem, Mona A. Esawy.
© 2026.
40 pages.
|
|
|