Deep learning-based image manipulation localization (IML) methods have achieved remarkable performance in recent years, but typically rely on large-scale pixel-level annotated datasets. To address the challenge of acquiring high-quality annotations, some recent weakly supervised methods utilize image-level labels to segment manipulated regions. However, the performance is still limited due to insufficient supervision signals. In this study, we explore a form of weak supervision that improves the annotation efficiency and detection performance, namely scribble annotation supervision. We re-annotated mainstream IML datasets with scribble labels and propose the first scribble-based IML (Sc-IML) dataset. Additionally, we propose the first scribble-based weakly supervised IML framework. Specifically, we employ self-supervised training with a structural consistency loss to encourage the model to produce consistent predictions under multi-scale and augmented inputs. In addition, we propose a prior-aware feature modulation module (PFMM) that adaptively integrates prior information from both manipulated and authentic regions for dynamic feature adjustment, further enhancing feature discriminability and prediction consistency in complex scenes. We also propose a gated adaptive fusion module (GAFM) that utilizes gating mechanisms to regulate information flow during feature fusion, guiding the model toward emphasizing potential tampered regions. Finally, we propose a confidence-aware entropy minimization loss. This loss dynamically regularizes predictions in weakly annotated or unlabeled regions based on model uncertainty, effectively suppressing unreliable predictions. Experimental results show that our method outperforms existing fully supervised approaches in terms of average performance both in-distribution and out-of-distribution.

Microseismic (MS) technology is one of the effective methods for engineering fracture monitoring and has been widely used in coal mining. The accurate P-phase arrivals are critical for improving the accuracy of source locations and MS parameters. In this study, a multistep method (CNN-KA) combining convolutional neural network (CNN) combined, K-means and Akaike information criterion (AIC) is proposed. First, CNN is applied to recognize mining-induced waveforms of rock fractures. Second, the maximum overlapping discrete wavelet transform and multiresolution analysis (MRAs) are integrated to denoise for acquired waveforms. Subsequently, a new picker was developed by introducing K-mean clustering to AIC. Finally, the performance of the proposed CNN-KA was evaluated with open access and in-field data. Combined with software development, the proposed method was applied in 12123 working face of Paner coal mine in Anhui Province, China. The results show that the accuracy of CNN-KA in recognizing mining-induced rock fracturing events is 98.11%. The mean absolute error (MAE) of CNN-KA is 0.0915 s at 200-Hz frequency, which is 86.65% lower than STA/LTA (short-term average to long-term average (STA/LTA), one of the most commonly used in MS technology). The errors of picking and positioning are 3.7 ms and 10.89 m, respectively, at 5000-Hz frequency. MS source parameters based on CNN-KA were effectively used to interpret the abnormal dynamic response of the roof stratum in 12123 working face of Paner coal mine, which was confirmed by real-time mine pressure and design data of the residual coal pillars. The results can be further used for calculating precise MS source parameters and early warning of engineering geohazards.