Advancing Microplastic Detection Technology through Digital Image Processing, Fractal Analysis, and Polynomial Approximation Methods

Microplastics, which are small solid polymer particles with size of ≤5 mm, are becoming a global threat to the environment. They result from more extensive plastic materials breakdown due to natural processes (weathering) or human activities (such as commercial use.) Because microplastics are found in an aquatic ecosystem, it is essential to have accurate detection methods. Detecting microplastics is crucial for understanding their impact on ecosystems and human health. This study proposes an integrated approach that combines Digital Image Processing (DIP), Fractal Dimension Analysis (FDA) and Polynomial Approximation to identify microplastic robustly. In our pursuit of analyzing environmental contaminants, we engineered a microscopy prototype designed to examine the environmental samples, including water and sediment. This prototype is distinguished by its integration of a portable microscope, enabling precise navigation across samples on the xy-plane while adeptly adjusting the focus along the z-axis through a bespoke mechanical setup. This innovation enhances the device's utility in microplastic detection and significantly broadens its applicability in field settings, allowing for on-site analysis with high precision. Our approach integrates manual adjustments to fine-tune the working distance, facilitating targeted, image capture. Using a hands-on image acquisition method, operators can manually trigger the photo-capture process upon identifying optimal viewing conditions; this ensures that each image, saved in JPG format. The primary objective of capturing these images is to identify and characterize microplastic particles across diverse environmental matrices. All images are acquired under meticulously controlled lighting conditions to ensure accuracy and consistency, thereby minimizing potential noise and artifacts that cloud compromise data quality. Our sample collection encompasses a variety of matrices, including water and sediments, chosen for their high likelihood of microplastic contamination. This enhanced methodology improves the precision of our environmental assessments and paves the way for a deeper understanding of microplastics distribution and its ecological impacts. In this study, DIP techniques were crucial in preprocessing the acquired images to facilitate microplastic detection. Initially, we focused on improving the visibility of microplastic within the images through contrast adjustments. Although histogram equalization was considered, we primarily enhanced contrast via adaptative stretching. To address the challenge of image noise-without compromising the integrity of microplastic edges-we implemented Gaussian and median filters, effectively reducing noise levels. Segmenting microplastic regions from the background was achieved using advanced thresholding techniques, including Otsu’s and adaptative thresholding, which isolated microplastic particles for further analysis. Following segmentation, our feature extraction approach involved analyzing the detected particles’ shape, texture, and color characteristics. Shape descriptors such as area, perimeter, Feret diameter, and circularity were calculated to provide a geometric profile of each particle, offering insights into their physical dimensions and overall morphology. Additionally, color attributes were assessed through RGB or HSV histograms, capturing the distinct color information of microplastic. This comprehensive processing and analysis pipeline allowed for a nuanced understanding of microplastic properties, facilitating more accurate identification and characterization of environmental samples. FDA and PA play crucial roles in assessing the complexity of microplastic shapes and contours. The complexity of microplastic shapes is quantified by calculating their fractal dimensions, utilizing methods such as box-counting. Higher fractal dimensions indicate the irregular and convoluted shapes typical of microplastics, providing numerical insight into their complexity. Simultaneously, the contours of microplastics are modeled using polynomial functions. This process involves fitting polynomials to the boundary points of particles, resulting in smooth approximations that capture the nuances of microplastic shapes. The coefficients of these polynomials serve as key discriminative features, aiding in the subsequent classification process. In the classification phase, patterns-including features derived from shape, fractal dimensions analysis, and polynomial approximation-are analyzed to identify potential microplastic. This groundwork facilitates using classifiers like Support Vector Machine (SVM), Random Forest (RF) or pattern recognition techniques (e.g., K-means clustering) in identifying microplastics, establishing a comprehensive method for distinguishing them in environmental samples. Results indicate that the combined approach enhances accuracy in microplastic detection over traditional methods. Integrating FDA and PA improves the discrimination between microplastic and other particles. This methodology allows for rapidly analyzing large image datasets, facilitating efficient processing. Additionally, it enables accurate quantification of microplastic counts and size distributions. Preliminary experiments on real-world samples have yielded promising results, highlighting the potential of this approach for widespread environmental monitoring and assessment. Challenges in this field arise from the variability in particle shape, size, and the complexity of the backgrounds against which these particles are identified. Additional challenges include the dynamic nature of environmental samples and the need for high-throughput analysis capabilities to handle the vast amounts of data generated. Future directions to enhance microplastic detection involve incorporating multispectral imaging to utilize additional spectral bands, thereby improving discriminations capabilities between microplastics and other materials. Furthermore, exploring deep learning, specifically convolutional neural networks (CNN), for end-to-end feature learning presents an opportunity to automate and refine the detection process. The proposed framework holds practical implications for environmental monitoring, waste management, and risk assessment, emphasizing the need for ongoing optimization of parameters, validation of results across diverse sample types, and the development of real-time detection systems. In conclusion, integrating DIP, FDA, and PA establishes a comprehensive framework for microplastic detection. To fully realize its potential, future research should focus on the optimization and validation of this approach across various conditions and the practical implementation of real-time detections systems, thereby enhancing the effectiveness and efficiency of environmental monitoring strategies. Extraction of the characteristics of the different materials and classification. Detection of microplastics using different techniques such as edge detection, filters, binarization, polynomials.