The paper can be accessed on ArXiv: https://arxiv.org/abs/2204.03343.
The weather dataset from NEA Singapore for the real-world experiments was retrived from https://data.gov.sg/search?groups=environment on 7 Jan 2022.
The Singapore boundary file (.shp) was retrieved from https://maps.princeton.edu/catalog/stanford-pg798kr1205 on 20 Jan 2022.
Multiple expriments have been conducted to validate our algorithms. In the synthetic_experiment.m, we create a synthetic dataset of sensor network depolyed in R^2 with 2500 spatial locations over the simplex [-5,5] x [-5,5]. At each sensor location, based on the value of the spatial field, a sequence of point/integral observations is generated from either H0 or H1.
Based on the sensor type, each sensor perfroms WGPLRT or NLRT to make the local inference, i.e. to classify H0/H1. Then, SBLUE is used to reconstruct the binary spatial field at locations where no sensors are placed.
In the semi_synthetic_experiment.m, the experiments are performed based on a real dataset from NEA Singapore.
To test the efficacy of individual algorithms, we also conduct experiments on WGPLRT, NLRT, SBLUE respectively. See test_WGPLRT.m,test_NLRT.m,test_SBLUE.m.
Please refer to the paper for the detailed results.
| No. | Description | Code |
|---|---|---|
| 1 | Synthetic Experiments | src |
| 1 | Real-world Experiments | src |
| 2 | Test for WGPLRT | src |
| 3 | Test for NLRT | src |
| 4 | Test for SBLUE | src |
During the implementation of the codes, we use the GPML package provided by Carl Edward Rasmussen & Hannes Nickisch, which is accessible on http://gaussianprocess.org/gpml/code/matlab/doc/.