- Research Projects
- >
- Nereid
Nereid - Using a Convolutional Neural Network (CNN) Approach, an AI Technique, to Rapidly and Accurately Detect Bacteria That Might Cause Water-Borne Diseases
Abstract:
According to WHO, 2.1 billion people lack access to clean-water, and approximately 1 out of 3 drinks from such heavily-contaminated water that they are at high-risk for severe water-borne diseases. The current-methods for microbe-detection time and cost-consuming. This proposal is a novel and interdisciplinary model using a neural network to detect microbial-presence in water. A cohesive-device was created with 3 separate systems. The first was a novel and cost-effective microscope-imaging system which utilized a microscopic lens attachment, a Raspberry Pi Zero and a Raspberry Pi camera. The second was the custom-trained neural network to analyze images of water contamination. This network was trained on 3 microbe types: Lactobacillus, Streptococcus, and Saccharomyces Cerevisiae, using TensorFlow. It used 2-features and the stochastic-gradient-descent algorithm. This training accuracy was consistently above 90%. The validation demonstrated an average accuracy of 98.53% and an F-Score of 95.70%. Lastly the transmission system used Long-Range Radio (LoRA) for communication between the “Water Pipe” and “Water Plant”. The end-to-end method is that the microscopic system takes images at regular intervals with a cron-job. These images are sent to the neural network to be analyzed for possible contamination. The network outputs the classification of the contamination, which is sent to the transmission system. If contamination is present, this system will send it to the “Water Plant” from the “Water Pipe” using the LoRA technologies. This solution is scalable and can be trained to detect multiple microbes and other contaminants and can be deployed as a commercial method of detection.
According to WHO, 2.1 billion people lack access to clean-water, and approximately 1 out of 3 drinks from such heavily-contaminated water that they are at high-risk for severe water-borne diseases. The current-methods for microbe-detection time and cost-consuming. This proposal is a novel and interdisciplinary model using a neural network to detect microbial-presence in water. A cohesive-device was created with 3 separate systems. The first was a novel and cost-effective microscope-imaging system which utilized a microscopic lens attachment, a Raspberry Pi Zero and a Raspberry Pi camera. The second was the custom-trained neural network to analyze images of water contamination. This network was trained on 3 microbe types: Lactobacillus, Streptococcus, and Saccharomyces Cerevisiae, using TensorFlow. It used 2-features and the stochastic-gradient-descent algorithm. This training accuracy was consistently above 90%. The validation demonstrated an average accuracy of 98.53% and an F-Score of 95.70%. Lastly the transmission system used Long-Range Radio (LoRA) for communication between the “Water Pipe” and “Water Plant”. The end-to-end method is that the microscopic system takes images at regular intervals with a cron-job. These images are sent to the neural network to be analyzed for possible contamination. The network outputs the classification of the contamination, which is sent to the transmission system. If contamination is present, this system will send it to the “Water Plant” from the “Water Pipe” using the LoRA technologies. This solution is scalable and can be trained to detect multiple microbes and other contaminants and can be deployed as a commercial method of detection.
