MACNeXt-Based Bacteria Species Detection

Abstract

Bacteria underpin human health, environmental balance, and industrial processes. Rapid and accurate identification is essential for diagnosis and responsible antibiotic use. Culture, biochemical tests, and microscopy are slow, expensive, and depend on expert judgment, which introduces subjectivity and errors. This research aims to recommend a new generation deep learning architecture for bacterial species classification. We curated a bacterial image dataset, and this dataset contains 18,221 microscopic images from 24 species under standard laboratory conditions. All images passed clarity and focus checks. We developed a compact CNN, the Multiple Activation Network (MACNeXt). The recommended MACNeXt preserves local feature extraction and improves representation with two activation functions (GELU and ReLU) and a multi-branch design. The aim is high accuracy with low computational cost for routine clinical use. MACNeXt achieved 90.97% accuracy, 89.63% precision, 88.64% recall, and 88.99% F1-score on the test set. The calculated results and findings showcase balanced and stable performance across species with an efficient, lightweight design since the introduced MACNeXt has about 4.4 million learnable parameters. The results of the MACNeXt openly demonstrate that this CNN is a compact, lightweight, and highly accurate CNN model.