Scholarlink

In an era characterized by the growing need for efficient image classification, this study investigates the performance of serial and parallel processing techniques for classifying fruit images as “Healthy” or “Rotten” using a pretrained ResNet18 neural network. The classification algorithm begins with preprocessing input images, which are then passed through the ResNet18 model for feature extraction and classification. The serial method processes images sequentially, while the parallel approach employs threading to simultaneously process image batches. A flowchart is presented to visually outline the workflow, including image loading, preprocessing, classification, and result aggregation. Detailed experiments compare the cumulative processing times of serial and parallel methods, demonstrating significant improvements in computational efficiency through parallelization. The results validate the proposed framework as a scalable and practical solution for large-scale, real-time image classification tasks, maintaining accuracy while reducing processing time.

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

Performance Comparison of Serial and Parallel Fruit Classification Using Pretrained Neural Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • M. Mallegowda,
  • A. Parkavi,
  • S. Sanath,
  • Siddharth Satyavolu,
  • Moulya R. Gowda

摘要

In an era characterized by the growing need for efficient image classification, this study investigates the performance of serial and parallel processing techniques for classifying fruit images as “Healthy” or “Rotten” using a pretrained ResNet18 neural network. The classification algorithm begins with preprocessing input images, which are then passed through the ResNet18 model for feature extraction and classification. The serial method processes images sequentially, while the parallel approach employs threading to simultaneously process image batches. A flowchart is presented to visually outline the workflow, including image loading, preprocessing, classification, and result aggregation. Detailed experiments compare the cumulative processing times of serial and parallel methods, demonstrating significant improvements in computational efficiency through parallelization. The results validate the proposed framework as a scalable and practical solution for large-scale, real-time image classification tasks, maintaining accuracy while reducing processing time.