Multiclass Boundary Refinement with Neural Cellular Automata for Remote Sensing Land Cover Segmentation

Pascoassis Souza Santos Meira; Gilson Antonio Giraldi; Rafael Duarte Coelho dos Santos

cs.CVOct 21, 2026

Multiclass Boundary Refinement with Neural Cellular Automata for Remote Sensing Land Cover Segmentation

Authors: Pascoassis Souza Santos Meira, Gilson Antonio Giraldi, Rafael Duarte Coelho dos Santos

Organizations: Instituto Nacional de Pesquisas Espaciais (INPE), São José dos Campos, SP, Brazil

Abstract

Semantic segmentation of remote sensing imagery often places class boundaries a few pixels away from where they truly lie, and this position error is concentrated in the thin band measured by Boundary IoU (BIoU). We treat boundary refinement as the iterated application of a small learned local rule: the Boundary Refinement Neural Cellular Automaton (BRNCA), that starts from the class probabilities of a frozen, pre-trained segmenter and corrects them cell by cell. We generalise the binary rNCA to the multiclass case by keeping the visible channels on the probability simplex through an Aitchison (log-space) update and by adding free hidden channels. The best configuration improves the mean BIoU over the base segmenter by 2.36 points (from 46.63 to 48.99). The gain is concentrated on the irregular woodland class (+8.5 points BIoU); this suggests that position error is correlated with the fractal dimension of the class boundaries.

Figures & tables

Figure 1: Position and semantic errors in a real segmentation. Left: input image; middle: error regions overlaid on it; right: error regions alone. Position errors are thin slivers along region contours; the large light-green blob is a semantic error (sparse vegetation predicted over an area of dense vegetation).

Panel of Figure 2 — Figure 2: One BRNCA step. The frozen segmenter provides the initial state once; the loop (solid arrows) perceives the state, computes an increment $\delta=(\delta^{p},\delta^{h})$ with a shallow MLP conditioned on the image, and updates every cell synchronously ( Equation 2 ).

Figure 3: Cropped example of a LandCover.ai aerial image and its annotation mask, with colour legend for the five segmentation classes.

Class	$D$	$\Delta$ IoU	$\Delta$ BIoU	$\Delta$ IIoU
water	1.129	$+0.01$	$+0.21$	$\phantom{-}0.00$
building	1.192	$+0.04$	$+0.39$	$-0.02$
road	1.260	$-0.05$	$+0.82$	$-0.04$
woodland	1.264	$+2.15$	$\mathbf{+8.49}$	$-0.79$
background	1.330	$+0.55$	$+1.89$	$-0.32$
Mean	1.235	$+0.54$	$+2.36$	$-0.23$

Table 2: Per-class change with respect to the base segmenter, LandCover.ai validation, best epoch, for the best configuration of the ablation by BmIoU (

m=15

d_{h}=256

; Table 1 ), in pp.

D

is the fractal dimension of the class boundary ( Section 4.1 ). Classes are sorted by increasing

D

; the last row is the unweighted mean over classes (

\Delta

mIoU,

\Delta

BmIoU,

\Delta

mIIoU).

Figure 4: Evolution of the automaton over its 64 steps on LandCover.ai. Panels are read column by column, left column first. Top left: input image and ground-truth label (yellow: background; blue: building; grey: road). The remaining panels show, for $t\in\{1,8,16,24,32,40,48,56,64\}$ , the predicted label (left) and the uncertainty map (right; dark: low, bright: high; defined by the entropy of $p_{t}$ ).

Figure 5: Gain of mean BIoU over the base segmenter as a function of the number of automaton steps, for the best configuration ( $m=15$ , $d_{h}=256$ ), run for twice the 64 steps used at inference. The gain peaks at 2.36 pp, in the plateau between roughly $t=60$ and $t=72$ , and falls to about 0.6 pp at $t=128$ .

Remote Sensing Image Segmentation

Remote sensing image segmentation labels land cover and objects in satellite, aerial, radar, and multispectral imagery.

See all 220 papers

Beyond Accuracy: Uncertainty-Guided Boundary Refinement for Reliable Biomedical Image Segmentation
Sep 11, 2026Anima KujurMedical Image Segmentation Boundary-Aware Segmentation
Interdisciplinary Centre for Scientific Computing (IWR), Heidelberg University, Heidelberg, Germany
BASeg: Boundary-Aware Remote Sensing Segmentation with Structural Penalties
Aug 16, 2026Yuexi Song, Kailai Sun, Zhuoyu Wang +4Remote Sensing Image Segmentation Boundary-Aware Segmentation
National University of Singapore · Singapore-MIT Alliance for Research and Technology · Massachusetts Institute of Technology +1
CROSS: Cascaded Distillation and Dual-Constraint Grounding for Remote Sensing Referring Segmentation
Aug 4, 2026Tingzhang Luo, Ruizhong Liu, Yichao Liu +3Referring Image Segmentation
City University of Hong Kong · The Hong Kong University of Science and Technology (Guangzhou) · Nankai University +1

Sep 11, 2026J/K move · Enter open · S save

Beyond Accuracy: Uncertainty-Guided Boundary Refinement for Reliable Biomedical Image Segmentation

Anima Kujur

Interdisciplinary Centre for Scientific Computing (IWR), Heidelberg University, Heidelberg, Germany

Medical Image Segmentation Boundary-Aware Segmentation Segmentation Uncertainty Estimation

Open paper AskPDF arXiv

Accurate biomedical image segmentation requires not only high global overlap but also reliable delineation of clinically meaningful boundaries. In blood-smear microscopy, cytoplasm and nucleus contours provide the structural basis for downstream morphology analysis; however, deep segmentation models may remain uncertain or overconfident near ambiguous boundary regions even when achieving strong Dice scores. This work proposes a Reliability-Aware Boundary Refinement Network (RABR-Net), a two-stage framework for trustworthy image segmentation. A strong UNet++ EfficientNet-B4 base segmenter first produces initial class probabilities and logits. Predictive entropy, test-time augmentation variance, margin uncertainty, probability gradients, and soft boundary cues are then combined into a boundary-aware reliability representation. This representation guides a gated residual refiner that selectively corrects uncertain boundary pixels while preserving confident regions of the base prediction. The framework is evaluated using overlap accuracy, class-wise Dice, Boundary Dice, HD95/ASSD, calibration, risk--coverage analysis, robustness under image perturbations, qualitative correction maps, and paired statistical testing. On the held-out test set, the proposed method improves Dice from 0.9602 to 0.9614, Boundary Dice from 0.3448 to 0.3611, and HD95 from 3.0354 to 2.8274 compared with the cached base prediction. Statistical analysis confirms significant improvements in Dice, Boundary Dice, and HD95. Qualitative results show that the learned gate concentrates around uncertain cytoplasm and nucleus boundaries, and correction maps confirm localized boundary refinement. Although calibration does not automatically improve after refinement, the proposed framework provides an interpretable and reliability-focused strategy for boundary-sensitive biomedical image segmentation.

Abstract

Remote Sensing Image Segmentation

Explore similar work

Beyond Accuracy: Uncertainty-Guided Boundary Refinement for Reliable Biomedical Image Segmentation

BASeg: Boundary-Aware Remote Sensing Segmentation with Structural Penalties

CROSS: Cascaded Distillation and Dual-Constraint Grounding for Remote Sensing Referring Segmentation