Research map / Time series / Foundation models for time series
GIST: Gear Type Identification by Spatiotemporal Trajectory Transformation for Monitoring Fisheries
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes GIST, a multi-stage pipeline that extracts fishing islands from AIS trajectories, converts SOG and ΔCOG distributions into 128×128 two-channel image-like representations, and classifies gear type (trawler, purse seiner, longliner) with a small CNN, reaching 97% accuracy.
Problem
Identifying fishing gear type from AIS vessel trajectories to support monitoring of IUU fishing, where existing sequence-based methods have uneven performance across gear types and trajectory lengths.
Approach
Use TripTracker to isolate fishing activity, then encode each fishing island's SOG and ΔCOG histograms over fixed-length segments as a fixed-size image, so a lightweight 2D CNN classifies gear type instead of an LSTM over variable-length time series; GANs (SDV PAR and CTGAN) augment the minority longliner class.
Key results
- GIST reaches 0.97 weighted accuracy and macro F1 0.97 across trawler, purse seiner and longliner classes, versus 0.78 for LSTM-seq and 0.95 for LSTM-img.
- GIST has 54,195 parameters and 107,850 FLOPs with 142 ms runtime, about four times fewer parameters and five times lower FLOPs than LSTM-img (214,851 params, 951 ms).
- Without data augmentation (GIST-w/oDA), longliner F1 drops to 0.53 versus 0.97 with augmentation.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| transform fishing islands into image-like representations and classify them with the CNN-based GIST pipeline | GIST · West Coast AIS 2017 test set | accuracy: 0.78 -> 0.97 | Helped |
| use GIST instead of LSTM-img | GIST · West Coast AIS 2017 test set | accuracy: 0.95 -> 0.97 | Helped |
| add data augmentation in the sequential modality (ablation) | GIST · Longliner | F1: 0.53 -> 0.97 | Helped |
| add data augmentation in the sequential modality (ablation) | GIST · West Coast AIS 2017 test set | accuracy: 0.95 -> 0.97 | Helped |
| augment longliner samples in the image-like modality rather than the sequential modality | GIST data augmentation · Longliner | F1: 0.97 -> 0.99 | Mixed |
| use GIST rather than LSTM-img | GIST · West Coast AIS 2017 test set | runtime (ms): 951 ± 8.02 -> 142 ± 1.84 | Helped |
Surprising findings
- Image-like augmentation with CTGAN gave slightly better results than sequential augmentation, yet the authors chose sequential-modality synthetic data to preserve generalizability because near-zero pixels in synthetic images introduced spurious features.
Limitations
- Only three gear types are classified; fixed gear and other fishing categories are excluded.
- Longliner class is very small (66 vessels, 43 native fishing islands), relying on synthetic augmentation.
- Evaluated only on West Coast U.S. AIS data from a single year (2017).
Experimental setup
Datasets: U.S. Coast Guard AIS data (West Coast, 2017) Global Fishing Watch Vessel Identity dataset
Benchmarks: AIS-derived gear type dataset (West Coast 2017)
Metrics: accuracy precision recall F1 parameter count FLOPs runtime
Baselines: LSTM-seq LSTM-img GIST-w/oDA
Builds on: TripTracker KerasTuner CTGAN PAR synthesizer
Closest papers by meaning
- Fish-Vista: A Multi-Purpose Dataset for Understanding & Identification of Traits from Images (CVPR 2025 · similarity 0.80)
- BirdCollect: A Comprehensive Benchmark for Analyzing Dense Bird Flock Attributes (AAAI 2024 · similarity 0.79)
- ReefNet: A Large-Scale Dataset and Benchmark for Fine-Grained Coral Reef Recognition (NeurIPS 2026 · similarity 0.78)
- Sequential Modeling of Complex Marine Navigation: Case Study on a Passenger Vessel (Student Abstract) (AAAI 2024 · similarity 0.78)
- SeafloorAI: A Large-scale Vision-Language Dataset for Seafloor Geological Survey (NeurIPS 2024 · similarity 0.78)
- WildlifeMapper: Aerial Image Analysis for Multi-Species Detection and Identification (CVPR 2024 · similarity 0.78)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.