Scholarlink

Bacterial species play a vital role in maintaining ecological balance by driving essential nutrient cycles. Understanding these microorganisms is crucial given their significance across various domains, including the food production, medical diagnostics, veterinary medicine, biogenetics, farming, pharmacology, and related disciplines. Nevertheless, the classification and identification of bacterial species pose significant challenges. The intricate similarities in their morphological characteristics make the process exceedingly complex, labor-intensive, and demanding of a structured methodology. To accurately identify and classify bacterial species, microbiologists must analyze both their phenotypic and genotypic traits. This process heavily depends on human expertise and requires high-cost equipment, making it resource-intensive. Automating this procedure through deep learning system for classification of bacterial images can reduce many problems faced by scientists in this field. Recent advancements in deep learning have demonstrated remarkable success in addressing complex problems, particularly in the domain of image identification. This research proposes a transfer learning-based deep learning method for classifying images of twenty different bacterial species: Acetobacter aceti, Acinetobacter baumannii, Alcaligenes faecalis, Arthrobacter methylotrophus, Bacillus anthracis, Bacillus circulans, Cytobacillus firmus, Deinococcus saxicola, Filibacter tadaridae, Halomonas piezotolerans, Janthinobacterium rivuli, Micrococcus luteus, Myroides, Nitratireductor soli, Priestia megaterium, Sphingomonas glacialis, Staphylococcus arlettae, Staphylococcus aureus, Streptomyces sp., and Thermus sp. The dataset consists of 2000 images, which were augmented by extracting eight patches from each image. For classification of these images, two CNN models MobileNetV2 and EfficientNetB3 were fine-tuned. This study conducted a comparison between these two different CNN models, and it was found that the MobileNetV2 model delivered better classification results than the EfficientNetB3.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Deep Learning-Based Approach to Microscopic Bacterial Species Classification

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shallu Kotwal,
  • Jatinder Manhas,
  • Tasleem Arif,
  • Vinod Sharma

摘要

Bacterial species play a vital role in maintaining ecological balance by driving essential nutrient cycles. Understanding these microorganisms is crucial given their significance across various domains, including the food production, medical diagnostics, veterinary medicine, biogenetics, farming, pharmacology, and related disciplines. Nevertheless, the classification and identification of bacterial species pose significant challenges. The intricate similarities in their morphological characteristics make the process exceedingly complex, labor-intensive, and demanding of a structured methodology. To accurately identify and classify bacterial species, microbiologists must analyze both their phenotypic and genotypic traits. This process heavily depends on human expertise and requires high-cost equipment, making it resource-intensive. Automating this procedure through deep learning system for classification of bacterial images can reduce many problems faced by scientists in this field. Recent advancements in deep learning have demonstrated remarkable success in addressing complex problems, particularly in the domain of image identification. This research proposes a transfer learning-based deep learning method for classifying images of twenty different bacterial species: Acetobacter aceti, Acinetobacter baumannii, Alcaligenes faecalis, Arthrobacter methylotrophus, Bacillus anthracis, Bacillus circulans, Cytobacillus firmus, Deinococcus saxicola, Filibacter tadaridae, Halomonas piezotolerans, Janthinobacterium rivuli, Micrococcus luteus, Myroides, Nitratireductor soli, Priestia megaterium, Sphingomonas glacialis, Staphylococcus arlettae, Staphylococcus aureus, Streptomyces sp., and Thermus sp. The dataset consists of 2000 images, which were augmented by extracting eight patches from each image. For classification of these images, two CNN models MobileNetV2 and EfficientNetB3 were fine-tuned. This study conducted a comparison between these two different CNN models, and it was found that the MobileNetV2 model delivered better classification results than the EfficientNetB3.