Authors: Pratiksha Chaudhari, University of Alabama
Title: Deep Learning for Real-Time Microplastic Detection in Environmental Samples
Abstract: Microplastics are plastic particles smaller than 5mm, including fibers, fragments, pellets, and other forms. Microplastics are present in almost all ecosystems and pose a serious threat to the environment and human health. While there are protocols for detecting microplastics in water samples, such as those set by the National Oceanic and Atmospheric Administration (NOAA) and the California Ocean Protection Council have been developed, these methods can be time-consuming, expensive, and require skilled personnel. To overcome these limitations, microplastic sensors based on computer vision, microfluidics, and deep learning have been developed. In recent research, machine learning algorithms like YOLOv4, RetinaNet, Faster R-CNN, SSD, Mask R-CNN, yolov3, etc., were used to detect microplastics in environmental samples. However, these methods have limitations in terms of accuracy, speed, and scalability. To address these limitations, the latest YOLOv5 deep learning model was used to detect microplastics. YOLOv5 showed better performance than previous versions and other algorithms. It includes a more powerful and efficient backbone called CSPNet, an improved learning strategy called self-labeling, an improved detection method, and a better post-processing method than any other algorithm. In addition, YOLOv5 is lightweight and can run on devices with limited processing power, making it ideal for real-time applications. Through optimization, YOLOv5 has shown significant improvements in recognition accuracy and speed. To train the YOLOv5 algorithm, a dataset of custom microplastic images was developed that includes a wide range of microplastic shapes, sizes, and colors commonly found in environmental samples. Data augmentation techniques were used to increase the diversity of the dataset and improve the model’s resilience to lighting and background changes. Model performance was evaluated using a dataset consisting of 10,000 environmental images with 27,000 microplastic annotations demonstrating the ability to detect microplastics of various sizes, shapes, and colors, including small and irregularly shaped microplastics. Quantitative measurements showed high accuracy and recall rates, indicating the ability to detect a large percentage of microplastics present in the video stream while minimizing false positives. To further enhance the detection of microplastics in environmental samples, we have used high-speed cameras in combination with the YOLOv5 deep learning model. This approach allows for the capture of rapid movement of microplastics in water samples, which can be difficult to detect with traditional methods. By incorporating high-speed cameras in our system, the YOLOv5 model can provide more accurate and efficient detection of microplastics, allowing for real-time monitoring and management of environmental pollution. Using YOLOv5 to detect microplastics has great potential for environmental monitoring and management, where real-time detection and identification of microplastics is essential for effective decision-making.