Slim Deep Learning Approach for Microplastics Image Classification in the Marine Environment

Microplastics are persistent pollutants in marine ecosystems, yet their accurate identification remains challenging due to their size, diverse morphologies, and the limitations of manual inspection methods. Deep learning offers improved accuracy for MP classification; however, most existing architectures, such as VGG16, U-Net, and ResNet, are computationally intensive for real-time or field-based applications. This study proposes a lightweight convolutional neural network, termed the Slim-DL-Model, designed to address the need for efficient, deployable microplastic classification. The model comprises three convolutional blocks followed by fully connected layers, optimized to reduce computational complexity while maintaining strong feature extraction capability. Using a dataset of holographic microplastic images representing polyethylene, polystyrene, and polyhydroxyalkanoate, the model was trained and evaluated using a 70/20/10 split. Results show rapid convergence and robust performance, achieving a precision of 0.9333 and an accuracy of 0.9761. With only 6.5 million trainable parameters, the Slim-DL-Model significantly reduces computational requirements compared to traditional deep architectures while maintaining competitive accuracy. The findings demonstrate the potential of the Slim-DL-Model for real-time, on-device microplastic classification in resource-limited coastal environments. This work presents a practical and scalable approach to automated microplastic monitoring, providing a foundation for future extensions that involve additional microplastic classes and edge–cloud integration.