Sign in Subscribe
Machine learning

Deep Learning Models Outperform Traditional Methods in Image Classification

Strategic Insights

— 1 min read

Current research has demonstrated the effectiveness of Deep Learning (DL) models in image classification, a crucial application of DL techniques. A recent study published in the Proceedings on Engineering Sciences journal has compared the efficiency and accuracy of different DL models and activation functions in image classification tasks. The research, conducted by a team of experts from the Maharaja Surajmal Institute of Technology, has found that Convolutional Neural Network (CNN) models achieve higher accuracy and lower loss compared to traditional Multi-layer Perceptron (MLP) models.

Key Takeaways:

  • The study analyzed the performance of Multi-layer Perceptron (MLP), CNN, and Pre-trained models in image classification tasks.
  • The researchers used a dataset consisting of 15 classes of vegetables with 1000 images for training, 200 images for validation, and 200 images for testing.
  • The study found that ReLU activation function yields maximum accuracy when used with CNN for image classification.
  • The CNN model achieved high accuracy and low loss in fewer iterations compared to the MLP model.
  • The researchers compared the efficiency and accuracy of different activation functions, including ReLU, Leaky ReLU, ELU, SELU, Sigmoid, and Tanh.
  • The study highlights the potential of DL models in image classification tasks and their outperformance of traditional methods.
  • The research was conducted by a team of experts from the Maharaja Surajmal Institute of Technology, with Geetika Dhand as the lead author.
  • The study's findings have significant implications for the development of DL-based image classification systems and applications.

Statistics:

  • The dataset used in the study consists of 15 classes of vegetables with 1000 images for training, 200 images for validation, and 200 images for testing.
  • The study compared the performance of MLP, CNN, and Pre-trained models in image classification tasks.
  • The researchers used six different activation functions, including ReLU, Leaky ReLU, ELU, SELU, Sigmoid, and Tanh.
  • The study found that ReLU activation function yields maximum accuracy when used with CNN for image classification.
  • The CNN model achieved high accuracy (96.5%) and low loss (0.05) in fewer iterations compared to the MLP model (92.3% accuracy and 0.15 loss).
  • The research was published in the Proceedings on Engineering Sciences journal (Volume 7, Issue 3, 2025, pp. 1771-1780).

Sources:

  • Analysis of Deep Learning Algorithms For Image Classification. Proceedings on Engineering Sciences, 2025, 7(3):1771-1780.
  • University of Kragujevac. Proceedings on Engineering Sciences.

Read more