Compact CNN-Based Bacteria Classification

What is CNN-based bacteria classification?

Clinical diagnostics, such as bacteria classification, including cancer and/or infectious diseases, are often based on microscopic images of biological samples. Enhancing the imaging system with a Convolutional Neural Network (CNN) automates the sample evaluation while also ensuring consistent quality and high-throughput screening.

In addition to individual components for imaging, Basler offers the development of a coordinated overall vision system, including image processing, analysis, and data transfer.

What is the challenge with bacterial classification?

Compact, fast, inexpensive, reliable – these are the increasingly important requirements for vision-based analytical devices for clinical diagnostics. Coordinated components from the lens, camera, cables, and processor board to the system, and CNN-based application software are crucial to solve these challenges. This is the only way to create a functioning vision system that meets the basic conditions for reliable results and diagnoses.

Bacteria classification method using a computer vision solution

Our case study identified four different bacterial species in the sample, and the result certainty is indicated by a percentage. The computer vision solution consists of hardware and software components that utilize Basler's camera solutions:

• A 1/1.8” S-mount lens

• Basler dart camera module with 5 MP resolution

• USB 3.0 interface and data cable

• A Jetson™ Nano processor board from NVIDIA®

The solution also includes both system and application software. The individual elements of the system software form a coherent system and can be easily integrated into a Linux architecture. As part of the application software, pylon AI Classification vTool performs image analysis based on the CNN. It analyzes the different types of bacteria and classifies them based on the classification model (CNNs) previously optimized with individual data sets. The CNN is trained on the host side, while the inference (image evaluation and result generation) takes place directly in the device using the pylon AI Classification vTool, which has been specially optimized for use on edge devices.

Benefits of Basler’s CNN-based bacteria classification imaging solutions

• Seamless, interoperable components across the entire vision system

• Hardware, software, and development from a single source

• Optimized image acquisition and simple integration of image analysis

• Flexible and interchangeable CNN

• Reliable and fast object classification with pylon AI Classification vTool