Classification Uncertainty Research Articles

A Probabilistic Filtering Approach to Non‐Effortful Responding

AbstractCommon response‐time‐based approaches for non‐effortful response behavior (NRB) in educational achievement tests filter responses that are associated with response times below some threshold. These approaches are, however, limited in that they require a binary decision on whether a response is classified as stemming from NRB; thus ignoring potential classification uncertainty in resulting parameter estimates. We developed a response‐time‐based probabilistic filtering procedure that overcomes this limitation. The procedure is rooted in the principles of multiple imputation. Instead of creating multiple plausible replacements of missing data, however, multiple data sets are created that represent plausible filtered response data. We propose two different approaches to filtering models, originating in different research traditions and conceptualizations of response‐time‐based identification of NRB. The first approach uses Gaussian mixture modeling to identify a response time subcomponent stemming from NRB. Plausible filtered data sets are created based on examinees' posterior probabilities of belonging to the NRB subcomponent. The second approach defines a plausible range of response time thresholds and creates plausible filtered data sets by drawing multiple response time thresholds from the defined range. We illustrate the workings of the proposed procedure as well as differences between the proposed filtering models based on both simulated data and empirical data from PISA 2018.

Quantifying uncertainty in land-use land-cover classification using conformal statistics

Land-use land-cover (LULC) change is one of the most important anthropogenic threats to biodiversity and ecosystems integrity. As a result, the systematic generation of annual regional, national, and global LULC map products derived from the classification of satellite imagery data have become critical inputs for multiple scientific disciplines. The importance of quantifying pixel-level uncertainty to improve the robustness of downstream analyses has long been acknowledged but this practice is still not widely adopted in the generation of these LULC products. The lack of uncertainty quantification is likely due to the fact that most approaches that have been put forward for this task are too computationally intensive for large-scale analysis (e.g., bootstrapping). In this article, we describe how conformal statistics can be used to quantify pixel-level uncertainty in a way that is not computationally intensive, is statistically rigorous despite relying on few assumptions, and can be used together with any classification algorithm that produces class probabilities. Our simulation results show how the size of the predictive sets created by conformal statistics can be used as an indicator of classification uncertainty at the pixel level. Our analysis based on data from the Brazilian Amazon reveals that both forest and water have high certainty whereas pasture and the “natural (other)” category have substantial uncertainty. This information can guide additional ground-truth data collection and the resulting raster combining the LULC classification with the uncertainty results can be used to communicate in a transparent way to downstream users which classified pixels have high or low uncertainty. Given the importance of systematic LULC maps and uncertainty quantification, we believe that this approach will find wide use in the remote sensing community.

Defect prediction of big data computer network based on deep learning model

Abstract Computer network software prediction is a good way to improve software quality, and the predictions of the software prediction method are close to the characteristics of the data set. In order to solve the problem that the invisible size of special data set is too large for computer software prediction, the author proposed a computer network-based software prediction method with deep computer coding and power learning. deep exploration of data features. data features. This type of model first uses an unsupervised learning-based evaluation model to evaluate the data set of 6 open projects, which solves the problem of classification uncertainty in the data; Deep self-encoding network models were then investigated. The model reduces the size of the data set, which is used to connect our model at the end of the model, the model uses training sets of shortened length to train the workers, and finally it makes predictions using the benchmarks. Experiments show that this model's prediction is better than standard software defect prediction, better than existing model-based software prediction models for fabric processes with repeated data volumes, and can be used in different categories. algorithms.

Flexible hidden Markov models for behaviour-dependent habitat selection

BackgroundThere is strong incentive to model behaviour-dependent habitat selection, as this can help delineate critical habitats for important life processes and reduce bias in model parameters. For this purpose, a two-stage modelling approach is often taken: (i) classify behaviours with a hidden Markov model (HMM), and (ii) fit a step selection function (SSF) to each subset of data. However, this approach does not properly account for the uncertainty in behavioural classification, nor does it allow states to depend on habitat selection. An alternative approach is to estimate both state switching and habitat selection in a single, integrated model called an HMM-SSF.MethodsWe build on this recent methodological work to make the HMM-SSF approach more efficient and general. We focus on writing the model as an HMM where the observation process is defined by an SSF, such that well-known inferential methods for HMMs can be used directly for parameter estimation and state classification. We extend the model to include covariates on the HMM transition probabilities, allowing for inferences into the temporal and individual-specific drivers of state switching. We demonstrate the method through an illustrative example of plains zebra (Equus quagga), including state estimation, and simulations to estimate a utilisation distribution.ResultsIn the zebra analysis, we identified two behavioural states, with clearly distinct patterns of movement and habitat selection (“encamped” and “exploratory”). In particular, although the zebra tended to prefer areas higher in grassland across both behavioural states, this selection was much stronger in the fast, directed exploratory state. We also found a clear diel cycle in behaviour, which indicated that zebras were more likely to be exploring in the morning and encamped in the evening.ConclusionsThis method can be used to analyse behaviour-specific habitat selection in a wide range of species and systems. A large suite of statistical extensions and tools developed for HMMs and SSFs can be applied directly to this integrated model, making it a very versatile framework to jointly learn about animal behaviour, habitat selection, and space use.

Theoretical characterization of uncertainty in high-dimensional linear classification

Being able to reliably assess not only the accuracy but also the uncertainty of models’ predictions is an important endeavor in modern machine learning. Even if the model generating the data and labels is known, computing the intrinsic uncertainty after learning the model from a limited number of samples amounts to sampling the corresponding posterior probability measure. Such sampling is computationally challenging in high-dimensional problems and theoretical results on heuristic uncertainty estimators in high-dimensions are thus scarce. In this manuscript, we characterize uncertainty for learning from a limited number of samples of high-dimensional Gaussian input data and labels generated by the probit model. In this setting, the Bayesian uncertainty (i.e. the posterior marginals) can be asymptotically obtained by the approximate message passing algorithm, bypassing the canonical but costly Monte Carlo sampling of the posterior. We then provide a closed-form formula for the joint statistics between the logistic classifier, the uncertainty of the statistically optimal Bayesian classifier and the ground-truth probit uncertainty. The formula allows us to investigate the calibration of the logistic classifier learning from a limited amount of samples. We discuss how over-confidence can be mitigated by appropriately regularizing.

Contribution of topographic features and categorization uncertainty for a tree species classification in the boreal biome of Northern Ontario

ABSTRACT Variations within local topography can effectively impact the location of tree species within naturally forested areas. Furthermore, the uncertainty of prediction for classification can vastly differ amongst topography and the overlying tree species groupings. This study investigated the supplementation of a suite of topographic features corresponding to morphometry and hydrological considerations, in addition to multispectral imagery and other LiDAR-derived features, at fine (2 m) spatial resolution for a pixel-based tree species classification of a forested region of the boreal biome in northern Ontario, Canada. The study area conforms to the Abitibi River Forest (ARF) and consists of the tree species of black spruce (Picea mariana), balsam fir (Abies balsamea), trembling aspen (Populus tremuloides), balsam poplar (Populus balsamifera), tamarack (Larix laricina), white spruce (Picea glauca), and eastern white cedar (Thuja occidentalis). Random forest (RF) and support vector machines (SVMs) were implemented for the classification. Topographic features, specifically those conforming to channel base level, valley depth, and multi-resolution valley bottom flatness (MRVBF), were among the most important features for species predictors. The RF and SVM methods were trained on pixels of pure stands (composed of 70%+ of same tree species) for the tree species groupings, which were split by site level. Modelling accuracies for both the pixel and site level were reported, with the best model attaining an overall site level accuracy and corresponding Cohen’s kappa score of 0.79 and 0.69 for classification, respectively. Entropy maps were generated to characterize the uncertainty of prediction, and substantiate that the regions of lowest uncertainty correspond to wetlands, which are dominated by black spruce (Picea mariana). A modified entropy map was calculated from the normalized top two probabilities of tree species groupings predicted per pixel, so as to better highlight regions of prediction uncertainty. A prediction map for the second most-likely tree species groupings was also computed, which supports the presence of balsam fir (Abies balsamea) as a secondary tree species throughout the ARF region.

Quantification of Surface Pattern Based on the Binary Terrain Structure in Mountainous Areas

Terrain significantly influences the physical processes and human activities occurring on the Earth’s surface, especially in mountainous areas. The classification and clarification of topographic structures are essential for the quantitative analysis of surface patterns. In this paper, we propose a new method based on the digital elevation model to classify the binary terrain structure. The slope accumulation is constructed to emphasize the accumulated topographic characteristics and is applied to support the segmenting process. The results show that this new method is efficient in increasing the completeness of the segmented results and reducing the classification uncertainty. We verify this method in three areas in South America, North America and Asia to evaluate the method’s robustness. Comparison experiments suggest that this new method outperforms the traditional method in areas with different landforms. In addition, quantitative indices are calculated based on the segmented results. The results indicate that the binary terrain structure benefits the understanding of surface patterns from the perspectives of topographic characteristics, category composition, object morphology and landform spatial distribution. We also assess the transferability of the proposed method, and the results suggest that this method is transferable to different digital elevation models. The proposed method can support the quantitative analysis of land resources, especially in mountainous areas and benefit land management.

A procedure for bridge visual inspections prioritisation in the context of preliminary risk assessment with limited information

Visual inspections represent the most resource-expensive operation in bridge management systems (BMSs). Consequently, their planning cannot be limited to the results of risk assessments, which are often performed under incomplete or uncertain information, but operating costs should also be accounted for. In this light, this paper presents a new methodology to prioritise visual inspections in BMSs based on risk condition and operating cost assessment. Framed in a context of limited information, the methodology exploits an information gain criterion to tackle potential uncertainties in the risk classification, so enabling to optimize the potential outcomes of the inspection plan. The proposed approach is conceived to be implemented in Geographical Information Systems (GISs) to facilitate the construction of intuitive risk maps and inspection plans. The effectiveness of the framework is demonstrated through its application to a simulated bridge stock whose risk condition was evaluated through the new guidelines for bridge risk classification and management recently adopted in Italy. The presented results highlight the benefits of simultaneously considering bridge risk conditions and inspection costs in defining the inspection plan, hence the usefulness of information theory to prioritise visual inspections. Limitations and future improvements of the proposed prioritisation method have also been highlighted and discussed.

Quantitative proteomic studies addressing unmet clinical needs in sarcoma.

Sarcoma is a rare and complex disease comprising over 80 malignant subtypes that is frequently characterized by poor prognosis. Challenges in clinical management include uncertainties in diagnosis and disease classification, limited prognostic and predictive biomarkers, incompletely understood disease heterogeneity among and within subtypes, lack of effective treatment options, and limited progress in identifying new drug targets and novel therapeutics. Proteomics refers to the study of the entire complement of proteins expressed in specific cells or tissues. Advances in proteomics have included the development of quantitative mass spectrometry (MS)-based technologies which enable analysis of large numbers of proteins with relatively high throughput, enabling proteomics to be studied on a scale that has not previously been possible. Cellular function is determined by the levels of various proteins and their interactions, so proteomics offers the possibility of new insights into cancer biology. Sarcoma proteomics therefore has the potential to address some of the key current challenges described above, but it is still in its infancy. This review covers key quantitative proteomic sarcoma studies with findings that pertain to clinical utility. Proteomic methodologies that have been applied to human sarcoma research are briefly described, including recent advances in MS-based proteomic technology. We highlight studies that illustrate how proteomics may aid diagnosis and improve disease classification by distinguishing sarcoma histologies and identify distinct profiles within histological subtypes which may aid understanding of disease heterogeneity. We also review studies where proteomics has been applied to identify prognostic, predictive and therapeutic biomarkers. These studies traverse a range of histological subtypes including chordoma, Ewing sarcoma, gastrointestinal stromal tumors, leiomyosarcoma, liposarcoma, malignant peripheral nerve sheath tumors, myxofibrosarcoma, rhabdomyosarcoma, synovial sarcoma, osteosarcoma, and undifferentiated pleomorphic sarcoma. Critical questions and unmet needs in sarcoma which can potentially be addressed with proteomics are outlined.

Quantifying amplitude-variation-with-offset uncertainties related to calcite-cemented beds using a Monte Carlo simulation

Calcite cement often occurs locally, forming thin layers of calcite-cemented sandstone characterized by high seismic velocities and densities. Because of their strong impedance contrast with the surrounding rock, calcite-cemented intervals produce detectable seismic reflection signals that may interfere with target reflections at the top of a reservoir. In this case, the amplitude-variation-with-offset (AVO) of the effective seismic signature will be altered and may even create a false hydrocarbon indication. From the Monte Carlo simulation, we find that the presence of thin calcite-cemented beds increases the uncertainty of the Bayesian pore-fluid classification based on the AVO attributes intercept and gradient. In the case example of a North Sea turbiditic oil and gas field, the probability of a false-positive hydrocarbon indication increases from 3%–5% to 18%–21% assuming an equal probability of the occurrence of brine, oil, and gas. The results confirm that calcite-cemented beds can create a pitfall in AVO analysis. Realistic estimates of the AVO uncertainty are crucial for the risk assessment of well placement decisions.

Analysis of the 24-h activity cycle: An illustration examining the association with cognitive function in the Adult Changes in Thought study.

The 24-h activity cycle (24HAC) is a new paradigm for studying activity behaviors in relation to health outcomes. This approach inherently captures the interrelatedness of the daily time spent in physical activity (PA), sedentary behavior (SB), and sleep. We describe three popular approaches for modeling outcome associations with the 24HAC exposure. We apply these approaches to assess an association with a cognitive outcome in a cohort of older adults, discuss statistical challenges, and provide guidance on interpretation and selecting an appropriate approach. We compare the use of the isotemporal substitution model (ISM), compositional data analysis (CoDA), and latent profile analysis (LPA) to analyze 24HAC. We illustrate each method by exploring cross-sectional associations with cognition in 1,034 older adults (Mean age = 77; Age range = 65-100; 55.8% female; 90% White) who were part of the Adult Changes in Thought (ACT) Activity Monitoring (ACT-AM) sub-study. PA and SB were assessed with thigh-worn activPAL accelerometers for 7-days. For each method, we fit a multivariable regression model to examine the cross-sectional association between the 24HAC and Cognitive Abilities Screening Instrument item response theory (CASI-IRT) score, adjusting for baseline characteristics. We highlight differences in assumptions and the scientific questions addressable by each approach. ISM is easiest to apply and interpret; however, the typical ISM assumes a linear association. CoDA uses an isometric log-ratio transformation to directly model the compositional exposure but can be more challenging to apply and interpret. LPA can serve as an exploratory analysis tool to classify individuals into groups with similar time-use patterns. Inference on associations of latent profiles with health outcomes need to account for the uncertainty of the LPA classifications, which is often ignored. Analyses using the three methods did not suggest that less time spent on SB and more in PA was associated with better cognitive function. The three standard analytical approaches for 24HAC each have advantages and limitations, and selection of the most appropriate method should be guided by the scientific questions of interest and applicability of each model's assumptions. Further research is needed into the health implications of the distinct 24HAC patterns identified in this cohort.

Associations of polygenic risk scores with posttraumatic stress symptom trajectories following combat deployment.

Identification of genetic risk factors may inform the prevention and treatment of posttraumatic stress disorder (PTSD). This study evaluates the associations of polygenic risk scores (PRS) with patterns of posttraumatic stress symptoms following combat deployment. US Army soldiers of European ancestry (n = 4900) provided genomic data and ratings of posttraumatic stress symptoms before and after deployment to Afghanistan in 2012. Latent growth mixture modeling was used to model posttraumatic stress symptom trajectories among participants who provided post-deployment data (n = 4353). Multinomial logistic regression models tested independent associations between trajectory membership and PRS for PTSD, major depressive disorder (MDD), schizophrenia, neuroticism, alcohol use disorder, and suicide attempt, controlling for age, sex, ancestry, and exposure to potentially traumatic events, and weighted to account for uncertainty in trajectory classification and missing data. Participants were classified into low-severity (77.2%), increasing-severity (10.5%), decreasing-severity (8.0%), and high-severity (4.3%) posttraumatic stress symptom trajectories. Standardized PTSD-PRS and MDD-PRS were associated with greater odds of membership in the high-severity v. low-severity trajectory [adjusted odds ratios and 95% confidence intervals, 1.23 (1.06-1.43) and 1.18 (1.02-1.37), respectively] and the increasing-severity v. low-severity trajectory [1.12 (1.01-1.25) and 1.16 (1.04-1.28), respectively]. Additionally, MDD-PRS was associated with greater odds of membership in the decreasing-severity v. low-severity trajectory [1.16 (1.03-1.31)]. No other associations were statistically significant. Higher polygenic risk for PTSD or MDD is associated with more severe posttraumatic stress symptom trajectories following combat deployment. PRS may help stratify at-risk individuals, enabling more precise targeting of treatment and prevention programs.

Effects of forest degradation classification on the uncertainty of aboveground carbon estimates in the Amazon

BackgroundTropical forests are critical for the global carbon budget, yet they have been threatened by deforestation and forest degradation by fire, selective logging, and fragmentation. Existing uncertainties on land cover classification and in biomass estimates hinder accurate attribution of carbon emissions to specific forest classes. In this study, we used textural metrics derived from PlanetScope images to implement a probabilistic classification framework to identify intact, logged and burned forests in three Amazonian sites. We also estimated biomass for these forest classes using airborne lidar and compared biomass uncertainties using the lidar-derived estimates only to biomass uncertainties considering the forest degradation classification as well.ResultsOur classification approach reached overall accuracy of 0.86, with accuracy at individual sites varying from 0.69 to 0.93. Logged forests showed variable biomass changes, while burned forests showed an average carbon loss of 35%. We found that including uncertainty in forest degradation classification significantly increased uncertainty and decreased estimates of mean carbon density in two of the three test sites.ConclusionsOur findings indicate that the attribution of biomass changes to forest degradation classes needs to account for the uncertainty in forest degradation classification. By combining very high-resolution images with lidar data, we could attribute carbon stock changes to specific pathways of forest degradation. This approach also allows quantifying uncertainties of carbon emissions associated with forest degradation through logging and fire. Both the attribution and uncertainty quantification provide critical information for national greenhouse gas inventories.

ECG Signal Analysis Using Convolution Neural Networks to avoid Uncertainty in Classification

This paper presents the study of electrocardiogram (ECG) signals analysis using convolution neural networks (CNNs) to avoid uncertainty in classification. MIT-BIH ECG dataset with five classes of beats, i.e., nonectopic, supraventricular, ventricular, fusion, unknown, is used for testing and training. The role of pre-processing of dataset was analyzed in improving the network efficiency. All the classes were balanced by doing under-sampling. The data was transformed by adding Gaussian noise to generalize training. CNN was designed with convolutional layer, max-pooling layer, concatenation layer and fully connected layer to classify the ECG signal. A dropout layer with the value of 0.4 was incorporated. Dropout layers are critical in CNN training because they minimize the training data from overfitting which leads to reduce the uncertainty in classification. Sigmoid activation function is used for eventual classification decision making. The presented network offers a high accuracy of 89.2%.

A Multimodal Data Fusion and Deep Learning Framework for Large-Scale Wildfire Surface Fuel Mapping

Accurate estimation of fuels is essential for wildland fire simulations as well as decision-making related to land management. Numerous research efforts have leveraged remote sensing and machine learning for classifying land cover and mapping forest vegetation species. In most cases that focused on surface fuel mapping, the spatial scale of interest was smaller than a few hundred square kilometers; thus, many small-scale site-specific models had to be created to cover the landscape at the national scale. The present work aims to develop a large-scale surface fuel identification model using a custom deep learning framework that can ingest multimodal data. Specifically, we use deep learning to extract information from multispectral signatures, high-resolution imagery, and biophysical climate and terrain data in a way that facilitates their end-to-end training on labeled data. A multi-layer neural network is used with spectral and biophysical data, and a convolutional neural network backbone is used to extract the visual features from high-resolution imagery. A Monte Carlo dropout mechanism was also devised to create a stochastic ensemble of models that can capture classification uncertainties while boosting the prediction performance. To train the system as a proof-of-concept, fuel pseudo-labels were created by a random geospatial sampling of existing fuel maps across California. Application results on independent test sets showed promising fuel identification performance with an overall accuracy ranging from 55% to 75%, depending on the level of granularity of the included fuel types. As expected, including the rare—and possibly less consequential—fuel types reduced the accuracy. On the other hand, the addition of high-resolution imagery improved classification performance at all levels.

On Measures of Uncertainty in Classification

On Measures of Uncertainty in Classification

How standardised are antibiotic regimens in otologic surgery?

BackgroundWithin otologic surgery, a paucity of well-controlled studies assessing the use of systemic antibiotic to reduce surgical site infections exists. Moreover, discrepancies in wound classification of procedures challenge consensus in antimicrobial prescribing patterns. We sought to compare surgeons from two different health systems to examine how surgeons’ prescribing habits compared to practice guidelines for numerous otologic procedures.MethodsAn online questionnaire was distributed to 33 Canadian and 32 Austrian surgeons who regularly perform otologic surgery. Current systemic antibiotic prescribing habits for cochlear implantation, cholesteatoma surgery, stapes surgery, and tympanoplasty ± ossiculoplasty were collected.ResultsEighteen of 33 (54.5%) Canadian surgeons provided responses, while 18 of 32 (56.3%) of Austrian surgeons answered. Clear consistency with clinical practice guidelines exists for pre-operative antibiotics use in cochlear implant surgery and infected cholesteatoma surgery. However, for stapes surgery and tympanoplasty ± ossiculoplasty, consensus is lacking for both pre- and post-operative antibiotic prescribing habits. Notable differences between the two countries include post-operative antibiotics for cochlear implant surgery (Austria: 36.4%, Canada: 71.4%) and uninfected cholesteatoma surgery (Austria: 33.3%, Canada: 77.8%). Across all procedures, both induction and post-operative antibiotic administration was not significantly associated with surgeon seniority when stratified by five-year increments.ConclusionThe lack of consensus among each country’s otologic surgeons underscores the uncertainty in wound classification and thus, adherence to clinical practice guidelines.Graphical

Addressing Uncertainty in Imbalanced Histopathology Image Classification of HER2 Breast Cancer: An Interpretable Ensemble Approach With Threshold Filtered Single Instance Evaluation (SIE)

Addressing Uncertainty in Imbalanced Histopathology Image Classification of HER2 Breast Cancer: An Interpretable Ensemble Approach With Threshold Filtered Single Instance Evaluation (SIE)

Bayesian Deep Learning for Hyperspectral Image Classification With Low Uncertainty

In recent years, deep learning models have been widely used for hyperspectral image (HSI) classification and most of existing deep learning-based methods merely focused on high classification accuracy. However, in real applications, classification with low uncertainty matters as much as accurate classification. Unfortunately, existing methods fail to consider uncertainty. To tackle this challenge, for the first time, Bayesian deep learning (BDL) is investigated to analyze the model uncertainty for HSI classification. Specifically, first, at the feature extraction stage, an HSI classification framework based on BDL, which contains two Bayesian Gabor layers and a global pooling layer (i.e., BDL-G <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ), is proposed. In BDL-G <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> , parameters in Gabor layers are sampled from the Gaussian distribution. The proposed BDL-G <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> not only provides the uncertainty estimation, but also strengthens the structure characteristic (i.e., texture) of HSI. Second, to model the uncertainty at the final classification stage, BDL-G <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> is combined with a Bayesian fully-connected layer (i.e., BDL-G <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -BFL), where the parameters’ distribution is adjusted adaptively. In the proposed BDL-G <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -BFL, the uncertainty at feature extraction and classification stages are both captured, and a whole uncertainty estimation framework is established. Experimental results on the three public HSI datasets demonstrates the superiority in both accuracy and uncertainty. The proposed Bayesian deep learning-based methods pioneer a new direction and provide useful inspiration and experience for practical applications.

Multiresolution-Based Rough Fuzzy Possibilistic C-Means Clustering Method for Land Cover Change Detection

Object-oriented change detection (OOCD) plays an important role in remote sensing change detection. Generally, most of current OOCD methods adopt the highest predicted probability to determine whether objects have changes. However, it ignores the fact that only parts of an object have changes, which will generate the uncertain classification information. To reduce the classification uncertainty, an improved rough-fuzzy possibilistic <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$c$</tex-math></inline-formula> -means clustering algorithm combined with multiresolution scales information (MRFPCM) is proposed. First, stacked bitemporal images are segmented using the multiresolution segmentation approach from coarse to fine scale. Second, objects at the coarsest scale are classified into changed, unchanged, and uncertain categories by the proposed MRFPCM. Third, all the changed and unchanged objects in previous scales are combined as training samples to classify the uncertain objects into new changed, unchanged, and uncertain objects. Finally, segmented objects are classified layer by layer based on the MRFPCM until there are no uncertain objects. The MRFPCM method is validated on three datasets with different land change complexity and compared with five widely used change detection methods. The experimental results demonstrate the effectiveness and stability of the proposed approach.