A Comparative Study of CNN Trainings for the Detection of E. Coli, P. Aeruginosa and S. Aureus: A Raspberry Pi-Based Prototype

Detection and classification of bacteria requires considerable time and highly skilled personnel. The use of convolutional neural networks (CNN), through deep learning and computer vision, is proposed for the automatic detection of E. Coli, P. Aeruginosa and S. Aureus. Using the spiral methodology, 36 iterations of the neural network were trained using RoboFlow and YOLOv5, where factors such as preprocessing, augmentation, illumination, camera angle and the number of classes to be classified were varied. The final dataset includes 75% of images from the AGAR Dataset and 25% of cultured samples from Yojoa Lake. Finally, a mAP of 92.2%, an accuracy of 88.8% and a recall of 91.8 % were obtained. This version of the CNN was used to make a prototype using 3D printing and SolidWorks, and a program using the Python programming language and a Raspberry Pi v4.