In China’s coastal regions, severe seawater eutrophication has led to frequent occurrences of red tides, causing significant damage to marine fisheries and aquatic resources. Therefore, red tide detection and prediction are of great research importance. Although current deep learning-based red tide detection methods perform well in detecting single-day red tides, they struggle with continuous multi-day detection due to insufficient mining of temporal features and difficulties in accurately capturing dynamic variations, limiting further improvements in detection accuracy. To address these issues, this study proposes a time-series fusion network model (CSF-RTDNet) for red tide detection using time-continuous GOCI data from the East China Sea. By integrating multi-temporal GOCI data, the model comprehensively captures spatiotemporal characteristics of red tides, enhancing dynamic process modeling. The CSF-RTDNet method improves feature discrimination by introducing NDVI to enhance red tide characteristics and increase separability between red tides and seawater. Additionally, an ECA channel attention mechanism is employed to fully exploit spectral features across different bands for deeper feature extraction. A novel feature extraction module, ASPC-DSC, combines atrous spatial pyramid convolution with depthwise separable convolution to effectively fuse multi-scale contextual features while improving computational efficiency. Furthermore, ConvLSTM is introduced to integrate temporal and spatial features, effectively addressing the insufficient mining of sequential characteristics in multi-day red tide detection. Experimental results demonstrate that CSF-RTDNet achieves robust detection of red tides with complex boundaries and continuous temporal patterns, attaining an accuracy of 95.89%, precision of 93.03%, recall of 96.34%, and a Kappa coefficient of 0.95. This method significantly enhances red tide detection accuracy and provides valuable technical support for marine environmental monitoring.

Reliable wall-relative localization is required for robotic cleaning of aquaculture-vessel bulkheads, where sensor availability changes across the air–water interface. A two-stage study separated measured dry-wall performance from a controlled evaluation of cross-medium estimation logic. In Stage I, five coordinate-estimation models (P0–P4) were evaluated at 23 non-initial control points on an 8 m × 4 m vertical steel wall. The hard-axis-constrained WO/IMU model P0 achieved an RMSE of 0.1806 m and a maximum error of 0.3818 m on the predominantly axis-aligned trajectories. In Stage II, five localization models (M1–M5) were evaluated over four prescribed scenarios and 400 paired software-in-the-loop realizations. All methods used the same simulated sensor streams and initial position; no reference-derived auxiliary velocity was supplied during estimation. Medium- and visual-quality adaptation reduced the 95th percentile of per-run maximum error from 0.5480 m for the fixed-parameter filter to 0.3756 m for M4 and reduced the 95th percentile of per-run maximum excess position jumps to approximately 0.151 m. M5 used encoder–IMU heading disagreement only to inflate process uncertainty and achieved an aggregate ATE-RMSE of 0.1024 m, which was practically indistinguishable from the 0.1028 m obtained by M4. Its remaining benefit was confined mainly to modest improvements in transition error and recovery behavior under prescribed slip. The results demonstrate estimator behavior in a measurement-informed feasibility study; synchronized tank and vessel experiments remain necessary.