We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Modified Adaptive Receptive Field Feature Extraction with Gated Fusion Adaptive Graph Neural Network for Microplastics Detection in Underwater Resources
Summary
A deep learning model combining progressive GAN preprocessing, Decoupled Faster R-CNN detection, and a Gated Fusion Adaptive Graph Neural Network classified microplastic images from underwater resources into fiber, film, fragment, and pellet types with 96.60% accuracy. This automated detection framework significantly advances scalable, reliable identification of microplastics in complex aquatic environmental samples.
Microplastics are tiny pieces of plastic waste that have accumulated worldwide in the location. Analyzing environmental sample images is difficult because the complexity of real data renders either automatic and manual analysis unreliable or time demanding. As a result, the proposed a spatial-temporal deep learning model Gated Fusion Adaptive Graph Neural Network (GFAGNN) is utilized to capture spatial features for detecting microplastics in underwater resources. Initially, microplastic images in water resources are gathered and then preprocessed using Progressive Wasserstein Generative Adversarial Network (PWGAN) and Edge-Aware Interactive Contrast Enhancement (e-IceNet) to reduce the noise and improve color improvement in microplastic images. After that, the next phase is object detection, the parts of a water image having microplastics are detected using a Decoupled Faster R-CNN - Multi-Scale Attention Mechanism (DeFRCNN-MAM). Using the Modified Adaptive Receptive Field Feature Extraction (MARFFE), extracts to detention the consistency of designs. The attributes are then input for Gated Fusion Adaptive Graph Neural Network (GFAGNN) in order to detect the image as "Fiber", "Film", "Fragment" or "Pallet" dependent on the plastics state. The suggested approach forecasts microplastics in underwater resources with an accuracy of 96.60%, a PPV of 94.20%, and FDR of 5.80%, correspondingly. Therefore, the suggested technique offers a thorough comprehension of all methods for identifying microplastics in order to decrease the possible impact on aquatic ecosystems.