|
This paper presents an algorithm-based change detection method for small-scale objects related to nuclear activities using geographic information system (GIS) data. From the nuclear nonproliferation perspective, the structural changes within the significantly suspected area for nuclear activities have to be captured. Additionally, the more amount of satellite imagery increases, e.g., CubeSats, the more systematic approach is required for change detection. Hence, the GIS vector data prescribed for the designated section is introduced as a guide layer in the process of change detection. It is supposed to stay up to date with a final interpretation to reflect the structure status for the next execution. The process of the proposed method consists of four steps: (1) Prior to change detection, satellite imagery of target areas is preprocessed, including the Gram-Schmidt pan-sharpening and image-to-image registration with a second-order rational polynomial coefficient (RPC) and nearest neighbour (NN) interpolation. (2) The before-and-after images obtained from the first step are analysed with the multivariate alteration detection (MAD), which produces pixel-based changes. The MAD output is imported as a change source layer in the process of change detection. (3) multi-temporal image object (MTIO) method is adopted for segmentation with all layers (4-band each and GIS vector layers). (4) The segmented layer stacks up behind the MAD layer to determine whether the MAD layer occupies over the threshold area in each segment, supported by the skeleton-based object linearity index (SOLI) and spectral homogeneity (SH) to minimise the shadow and building-lean effects. The Python programming language customised the MAD analysis, and the ENVI and eCognition support the rest process. The performance of the proposed method is reviewed with pixel-based accuracy assessment (precision, recall, and F1-score), and object-based criterion is also discussed in support of interpretation.
|
