Texture-Sensitive Superpixeling and Adaptive Thresholding for Effective Segmentation of Sea Ice Floes in High-Resolution Optical Images

Abstract

Efficient and accurate segmentation of sea ice floes from high-resolution optical (HRO) remote sensing images is crucial for understanding of sea ice evolutions and climate changes, especially in coping with the large data volume. Existing methods suffer from noise interference and mixture of water and ice caused high segmentation error and less robustness. In this study, we propose a novel sea ice floe segmentation algorithm from HRO images based on texture-sensitive superpixeling and twostage thresholding. First, sparse components are extracted from the HRO images using the Robust Principal Component Analysis (RPCA), and noise is removed by the bilateral filter. The enhanced image is obtained by combining the low-rank matrix and the sparse components. Second, a texture-sensitive Simple Linear Iterative Clustering (SLIC) superpixel algorithm is introduced for pre-segmentation of the enhanced HRO image. Third, a learning based adaptive thresholding in the two-stages is employed to generate the refined segmentation from the derived superpixels blocks. The efficacy of the proposed method is validated on two HRO images using visual assessment, quantitative evaluation (with seven metrics) and histogram comparison. The superior performance of the proposed method has demonstrated its efficacy for sea ice floe segmentation.

Electrochemical Intelligent Recognition of Mineral Materials Based on Superpixel Image Segmentation

“…Histogram is also an important basis for adjusting image contrast. In the digital age, histogram can be said to be everywhere [ 19 ]. We can use the slender peak of the image histogram to judge which gray level the main information of the image is concentrated on, use the area between the two peaks of the image histogram to judge which gray level the background color is on, or use the peak valley to judge the gray level of the noise.…”

Superpixel Based Sea Ice Segmentation with High-Resolution Optical Images: Analysis and Evaluation

“…Next, superpixel segmentation is generated. In this study, we have employed four algorithms for comparison, including the SLIC [13], Texture Sensitive SLIC (TS-SLIC) [8], Water Pixel (WP) [14], and Bayesian Adaptive Superpixel Segmentation (BASS) [15]. Next, k-means is applied to split the superpixels into two classes, namely water and ice, based on the mean intensity and standard deviation of each individual superpixel.…”

“…so called ice edges [5], and accurately determining the floe size distribution (FSD). Recent years, as the HRO images have become more accessible, various approaches have been developed for sea ice segmentation, including the combination of the maximum cross-correlation techniques and multi-sensor HRO images [6], the combination of the watershed transformation and random forest classification [7], and the texture-sensitive superpixel based segmentation [8]. To tackle the high computation complexity of the HRO imagery, superpixel algorithms provide an alternative representation of regular pixel grid by grouping the similar pixels into oversegmented small regions.…”

Towards a manual-free labelling approach for deep learning-based ice floe instance segmentation in airborne and high-resolution optical satellite images

“…Training a DL model requires a sufficiently annotated dataset. However, data labelling usually involves a lot of manual work and is expensive and time-consuming, which limits the application of DL methods to extracting individual ice floes (Jing and Tian, 2021;Zhou, 2017;Chai et al, 2020). In order to minimise the manual labelling effort required from the domain experts, we use classical image processing method to enable a manual-label free annotation of the dataset and automatically generate pseudo ground truth.…”