FakeReportNet: A Multimodal Deep Learning Approach for Detecting Fake Coastal Pollution Reports

Citizen-submitted reports of oil spills, plastic litter, and industrial discharge have become an important early-warning channel for coastal environmental monitoring, but the open submission pathways that make crowdsourced reporting fast and inexpensive also make it vulnerable to fabricated, manipulated, or misattributed content. This paper reviews fifteen published studies spanning multimodal fake-news detection, deepfake and image-forgery detection, deep-learning-based marine pollution detection, and AI-assisted citizen reporting, and proposes FakeReportNet, a conceptual multimodal deep learning framework for verifying the authenticity of citizen-submitted coastal pollution reports. The framework combines image forensics, textual credibility scoring, geotemporal consistency checking, and a fusion-level trust classifier that produces an explainable trust score for each report. The reviewed systems report strong results within their own domains, including 94.54% accuracy for a CNN-based deepfake classifier and 92.56% average precision for SAR-based oil-spill detection, yet none jointly verifies image, text, and metadata authenticity for coastal pollution reports. The paper outlines the proposed architecture, an evaluation protocol, and priority directions for implementation.