Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Compact waste image classification with multi-student CNNs and edge-oriented model selection.

Mohamed Echchidmi1, Anas Bouayad1

  • 1Laboratory of Artificial Intelligence, Data Science, and Emerging Systems, Sidi Mohamed Ben Abdellah University, Fez, Morocco.

Frontiers in Artificial Intelligence
|July 15, 2026
PubMed
Summary

This study introduces compact convolutional neural network (CNN) models for efficient waste classification on edge devices. LCNet-0.5 achieves high accuracy with minimal parameters, enabling practical smart-bin systems.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Cheatomaly: weakly supervised video anomaly ranking for exam cheating detection using vision transformers.

Frontiers in big data·2026
See all related articles

Area of Science:

  • Computer Vision
  • Machine Learning
  • Environmental Technology

Background:

  • Automatic waste classification is crucial for smart cities and recycling initiatives.
  • Current deep learning models face deployment challenges on resource-constrained edge devices due to size and latency.

Purpose of the Study:

  • To develop and evaluate edge-optimized compact convolutional neural network (CNN) frameworks for waste image classification.
  • To compare the accuracy-efficiency trade-offs of lightweight CNN architectures.

Main Methods:

  • A MobileNetV4 reference model was used to evaluate three lightweight student architectures: EfficientNet-Lite0, LCNet-0.5, and MobileNetV3-Small-0.5.
  • Models were trained and evaluated on the TrashNet dataset using a unified protocol, including stratified splits and five-fold cross-validation.
Keywords:
TinyMLedge computingknowledge distillationmodel compressionsustainabilitywaste classification

Related Experiment Videos

  • Knowledge distillation was explored to enhance student model performance.

    • Main Results:

    • MobileNetV4 achieved 97.09% top-1 accuracy.
    • LCNet-0.5 demonstrated strong performance (94.18% accuracy) with only 0.61 million parameters.
    • The LCNet-0.5 model was successfully integrated into a Raspberry Pi prototype, achieving ~1.0s inference latency and validated on the RealWaste dataset.

    Conclusions:

    • Lightweight architecture selection, informed by knowledge distillation and edge validation, is key for deployable waste classifiers.
    • LCNet-0.5 is a highly suitable sub-megabyte candidate for resource-constrained waste classification applications.
    • The developed framework enables accurate and efficient waste image classification on edge devices.