Evaluation of Global-Scale and Local-Scale Optimized Segmentation Algorithms in GEOBIA With SAM on Land Use and Land Cover

Abstract

Segmentation is crucial in geographic object-based image analysis (GEOBIA) for accurate land use and land cover (LULC) mapping. However, obtaining outstanding segmentation results in all scenarios proves challenging with a single algorithm. This study investigates seven segmentation algorithms: MF, SPRING, ESP2 (three global-scale algorithms), IODA, SA, EIODA (three local-scale optimization algorithms) and Segment Anything Model (SAM) (deep learning). In the custom dataset and semantic segmentation datasets, we apply visual interpretation, unsupervised, and supervised evaluation methods with 15 test images, using a total of 17 evaluation indices to assess the segmentation results. Based on the evaluation results, the effectiveness and adaptability of the algorithms in scene segmentation are comprehensively analyzed. The results report that global-scale segmentation approaches encounter difficulties in distinguishing meaningful objects in complicated scenarios. Both MF and SPRING methods are prone to over-segmentation. In many cases, ESP2 tends to generate homogeneous segments (low weighted variance (WV)), while EIODA tends to produce heterogeneous adjacent segments (low Moran's I (MI)). ED3 and segmentation evaluation index (SEI) demonstrate that SA and IODA can to some extent identify geo-objects, with SA being more effective and performing exceptionally well in building extraction. The EIODA performs well in areas with clear boundaries, like aquaculture ponds and water-land transitions. SAM accurately detects objects of various sizes, displaying rich semantic content and high consistency with reference polygons. The average intersection over union (IoU) reaches 71.10% and F measure attains 0.77 under normal conditions. 

Unsupervised Image Segmentation Parameters Evaluation for Urban Land Use/Land Cover Applications

“…This study proposed an unsupervised strategy for the selection of optimal segmentation-scale parameters and color factor weights. The new strategy presents alternative image segmentation evaluation paradigms in addition to those presented in the works of [5,16,25,[41][42][43][44]. The proposed robust, repeatable strategy for optimal selection of segmentation parameters (OSSP) consists of three modules.…”

“…However, in terms of the magnitude of the error, our proposed strategy achieved the best results at 0.006. The largest under-segmentation errors were achieved through the segmentation parameter selection methods proposed by [5,24,[42][43][44], respectively. The parameters evaluation strategy proposed by [41] achieved the lowest under-segmentation error at 0.005, followed by our proposed strategy at 0.008.…”

“…In terms of overall performance, the last column confirms that the newly proposed segmentation parameter evaluation strategy produced the lowest over and under-segmentation errors. [24] 0.257700 -------0.220500 0.127550 0.180000 El-Naggar, (2018) [42] 0.386800 −1.062000 0.329000 0.058000 0.273000 Vamsee et al, (2018) [43] 0.306700 −0.145000 0.311700 0.295000 0.368000 Wang et al, (2019) [5] 0.314500 -------0.148200 0.191500 0.171000 Norman et al, (2020) [41] 0.483200 0.004580 0.005000 0.009730 0.014000 Dao et al, (2021) [16] 0.490000 -------0.080000 0.970000 0.688000 He et al, (2024) [44] 0.321100 -------0.238000 0.246900 0.342900 The proposed method 0.003910 −0.005740 0.007780 0.007820 0.005510…”