Model Complexity Research Articles

Efficient Point-spread Function Modeling with ShOpt.jl: A Point-spread Function Benchmarking Study with JWST NIRCam Imaging

With their high angular resolutions of 30–100 mas, large fields of view, and complex optical systems, imagers on next-generation optical/near-infrared space observatories, such as the Near-Infrared Camera (NIRCam) on the James Webb Space Telescope, present new opportunities for science and also new challenges for empirical point-spread function (PSF) characterization. In this context, we introduce ShOpt, a new PSF fitting tool developed in Julia and designed to bridge the advanced features of PSFs in the full field of view (PIFF) with the computational efficiency of PSF Extractor (PSFEx). Along with ShOpt, we propose a suite of nonparametric statistics suitable for evaluating PSF fit quality in space-based imaging. Our study benchmarks ShOpt against the established PSF fitters PSFEx and PIFF using real and simulated COSMOS-Web Survey imaging. We assess their respective PSF model fidelity with our proposed diagnostic statistics and investigate their computational efficiencies, focusing on their processing speed relative to the complexity and size of the PSF models. We find that ShOpt can already achieve PSF model fidelity comparable to PSFEx and PIFF while maintaining competitive processing speeds, constructing PSF models for large NIRCam mosaics within minutes.

Surrogate model-based cognitive digital twin for smart remote maintenance of fusion reactor: modeling and implementation

Abstract A remote maintenance robot system (RMRS) plays a critical role in safeguarding the fusion energy experimental device’s security and stability. State-of-the-art intelligent technology such as cognitive digital twins (CDT) is widely considered capable of improving complex equipment’s performance and reducing management burden using a visualized system. However, the CDT virtual space cannot mirror the RMRS which is a kind of flexible multi-body system in real-time and with high fidelity. Therefore, we propose a cognitive digital twin (CDT) modeling method based on a surrogate model for the RMRS. Firstly, model-based system engineering (MBSE) is leveraged to build a structural modular architecture, which can decrease the modeling complexity of CDT and increase the modeling efficiency. Then, the surrogate models are self-learning within the CDT physical space, which reconstructs the RMRS’s real-time dynamic performances and endows CDT with cognitive capabilities. Finally, after integrating the CDT system, a smart decision-making plan that compensates for the operation error is generated for RMRS’s accurate control. We take a China Fusion Engineering Test Reactor (CFETR) multi-purpose overload robot (CMOR) as an example to demonstrate the implementation process. According to the results, CDT can achieve real-time (230 ms time delay) high-fidelity (5 mm control error) monitoring and accurate control, and CMOR conducts smart maintenance based on the simulation results. This method improves the efficiency of RM and provides solutions for high-duty cycle time of CFETR, it can also be applied to other tokamak fusion energy devices.

Schwarz decomposition for parallel minimum lap-time problems: evaluating against ADMM

The Minimum Lap Time Problem (MLTP) remains a significant area of research, particularly in the motorsport context. This form of Optimal Control Problem (OCP) aims to minimise lap times on a specific track with a given vehicle. Various complexities in both vehicle and track models are employed across the literature to address optimal trajectory planning. While previous works have tackled MLTP as a singular task using a serial approach, the increasing model complexity and horizon length demands the utilisation of parallelisation techniques. This paper introduces a novel application of the Overlapping Schwarz Decomposition algorithm to address the MLTP. The algorithm divides the problem into smaller sub-problems based on different sectors of a track, distributing them among multiple processors. We validate and compare the Schwarz approach against a serial approach and the Alternating Direction Method of Multipliers (ADMM) in solving MLTP with over 2.5 million variables. Despite the general efficiency improvement of parallelisation compared to the serial approach, the Schwarz algorithm demonstrates superior speed, accuracy and robustness compared to ADMM. As a result of our findings, it emerges as the preferred choice when large-scale MLTPs need to be solved.

Multi-scale collaborative modeling and deep learning-based thermal prediction for air-cooled data centers: An innovative insight for thermal management

Investigating the data center (DC) thermal environment and temperature distribution is crucial to responding to unforeseen events such as equipment failure or environmental changes. However, building full-scale simulation models from DC room level to chip level faces significant challenges. In this paper, we propose a distinctive approach that combines multi-scale collaborative modeling with deep learning techniques for thermal prediction in air-cooled DCs. By taking the simulation results of the parent model as the boundary conditions of the child model, we constructed the DC multi-scale simulation model, which significantly reduces the model complexity and computational resources. Leveraging experimental data, the models at different scales were validated separately. The effects of different cooling strategies, air supply temperatures and air supply flow rates on multi-scale simulation models were investigated. Based on the parametric simulation approach, datasets for training data-driven models are constructed. Simultaneously, we propose the CNN-BiLSTM-Attention neural network model to predict the maximum CPU temperature and optimize the hyperparameters of the neural network through by Bayesian optimization. The prediction results of the coupled multi-scale model and the deep learning prediction model show that the absolute error is controlled within ±0.1 K, and the R2 value of the model evaluation metric is as high as 0.9899. Herein, the results provide valuable insights for enhancing thermal management in air-cooled DCs, paving the way for more efficient and resilient DC operations in the future.

Research on a Method for Identification of Peanut Pests and Diseases Based on a Lightweight LSCDNet Model.

Timely and accurate identification of peanut pests and diseases, coupled with effective countermeasures, is pivotal for ensuring high-quality and efficient peanut production. Despite the prevalence of pests and diseases in peanut cultivation, challenges such as minute disease spots, the elusive nature of pests, and intricate environmental conditions often lead to diminished identification accuracy and efficiency. Moreover, continuous monitoring of peanut health in real-world agricultural settings demands solutions that are computationally efficient. Traditional deep learning models often require substantial computational resources, limiting their practical applicability. In response to these challenges, we introduce LSCDNet (Lightweight Sandglass and Coordinate Attention Network), a streamlined model derived from DenseNet. LSCDNet preserves only the transition layers to reduce feature map dimensionality, simplifying the model's complexity. The inclusion of a sandglass block bolsters features extraction capabilities, mitigating potential information loss due to dimensionality reduction. Additionally, the incorporation of coordinate attention addresses issues related to positional information loss during feature extraction. Experimental results showcase that LSCDNet achieved impressive metrics with accuracy, precision, recall, and Fl score of 96.67, 98.05, 95.56, and 96.79%, respectively, while maintaining a compact parameter count of merely 0.59 million. When compared with established models such as MobileNetV1, MobileNetV2, NASNetMobile, DenseNet-121, InceptionV3, and X-ception, LSCDNet outperformed with accuracy gains of 2.65, 4.87, 8.71, 5.04, 6.32, and 8.2%, respectively, accompanied by substantially fewer parameters. Lastly, we deployed the LSCDNet model on Raspberry Pi for practical testing and application and achieved an average recognition accuracy of 85.36%, thereby meeting real-world operational requirements.

Enhancing Deep Learning-Based City-Wide Traffic Prediction Pipelines Through Complexity Analysis

Deep learning models can effectively capture the non-linear spatiotemporal dynamics of city-wide traffic forecasting. Evidence of varying deep learning model performance between different cities, different prediction horizons, different scales, specific city regions, and during particular hours of the day abounds in the literature on deep learning-based traffic prediction, yet a unified metric to quantify the complexity of different prediction tasks does not exist. This paper proposes two metrics—model complexity (MC) and intrinsic complexity (IC). While MC quantifies the effective complexity of deep learning models for city-wide traffic prediction tasks, the IC quantifies the underlying complexity of the prediction task. Being an effective complexity metric, MC depends on the model and the data. The IC depends only on the data and is invariant to the model being used. Both metrics are validated through systematic experimentation using traffic volume data from three cities. Finally, we demonstrate how these metrics can improve the workflows for deep learning-based data-driven traffic prediction pipelines and deployment by reducing the hyperparameter search scope and comparing the effectiveness of different design pathways.

Estimation of Freshwater Discharge from the Gulf of Alaska Drainage Basins

The freshwater discharge from catchments along the Gulf of Alaska, termed Alaska discharge, is characterized by significant quantity and variability. Owing to subarctic climate and mountainous topography, the Alaska discharge variations may deliver possible impacts beyond the local hydrology. While short-term and local discharge estimation has been frequently realized, a longer time span and a discussion on cascading impacts remain unexplored in this area. In this study, the Alaska discharge during 1982–2022 is estimated using the Soil and Water Assessment Tool (SWAT). The adequate balance between the model complexity and the functional efficiency of SWAT suits the objective well, and discharge simulation is successfully conducted after customization in melting calculations and careful calibrations. During 1982−2022, the Alaska discharge is estimated to be 14,396 ± 819 m3⋅s−1⋅yr−1, with meltwater contributing approximately 53%. Regarding variation in the Alaska discharge, the interannual change is found to be negatively correlated with sea surface salinity anomalies in the Alaska Stream, while the decadal change positively correlates with the North Pacific Gyre Oscillation, with reasonable time lags in both cases. These new findings provide insights into the relationship between local hydrology and regional climate in this area. More importantly, we provide rare evidence that variation in freshwater discharge may affect properties beyond the local hydrology.

Complexity and enumeration in models of genome rearrangement

In this paper, we examine the computational complexity of enumeration in certain genome rearrangement models. We first show that the Pairwise Rearrangement problem in the Single Cut-and-Join model (Bergeron et al., 2010 [8]) is #P-complete under polynomial-time Turing reductions. Next, we show that in the Single Cut or Join model (Feijão and Meidanis, 2011 [21]), the problem of enumerating all medians (▪) is logspace-computable (FL), improving upon the previous polynomial-time (FP) bound of Miklós & Smith [41].

Lightweight Bearing Fault Diagnosis Method Based on Improved Residual Network

A lightweight bearing fault detection approach based on an improved residual network is presented to solve the shortcomings of previous fault diagnostic methods, such as inadequate feature extraction and an excessive computational cost due to high model complexity. First, the raw data are turned into a time–frequency map using the continuous wavelet transform, which captures all of the signal’s time- and frequency-domain properties. Second, an improved residual network model was built, which incorporates the criss-cross attention mechanism and depth-separable convolution into the residual network structure to realize the important distinction of the extracted features and reduce computational resources while ensuring diagnostic accuracy; simultaneously, the Meta-Acon activation function was introduced to improve the network’s self-adaptive characterization ability. The study findings indicate that the suggested approach had a 99.95% accuracy rate and a floating point computational complexity of 0.53 GF. Compared with other networks, it had greater fault detection accuracy and stronger generalization ability, and it could perform high-precision fault diagnostic jobs due to its lower complexity.

Visual Feature-Guided Diamond Convolutional Network for Finger Vein Recognition.

Finger vein (FV) biometrics have garnered considerable attention due to their inherent non-contact nature and high security, exhibiting tremendous potential in identity authentication and beyond. Nevertheless, challenges pertaining to the scarcity of training data and inconsistent image quality continue to impede the effectiveness of finger vein recognition (FVR) systems. To tackle these challenges, we introduce the visual feature-guided diamond convolutional network (dubbed 'VF-DCN'), a uniquely configured multi-scale and multi-orientation convolutional neural network. The VF-DCN showcases three pivotal innovations: Firstly, it meticulously tunes the convolutional kernels through multi-scale Log-Gabor filters. Secondly, it implements a distinctive diamond-shaped convolutional kernel architecture inspired by human visual perception. This design intelligently allocates more orientational filters to medium scales, which inherently carry richer information. In contrast, at extreme scales, the use of orientational filters is minimized to simulate the natural blurring of objects at extreme focal lengths. Thirdly, the network boasts a deliberate three-layer configuration and fully unsupervised training process, prioritizing simplicity and optimal performance. Extensive experiments are conducted on four FV databases, including MMCBNU_6000, FV_USM, HKPU, and ZSC_FV. The experimental results reveal that VF-DCN achieves remarkable improvement with equal error rates (EERs) of 0.17%, 0.19%, 2.11%, and 0.65%, respectively, and Accuracy Rates (ACC) of 100%, 99.97%, 98.92%, and 99.36%, respectively. These results indicate that, compared with some existing FVR approaches, the proposed VF-DCN not only achieves notable recognition accuracy but also shows fewer number of parameters and lower model complexity. Moreover, VF-DCN exhibits superior robustness across diverse FV databases.

Scheduling of memory chips for final testing on parallel machines considering power constraints and deteriorating effects

This paper delves into the intricate scheduling strategies crucial for the final testing phase of memory chip manufacturing within the semiconductor industry and other related sectors. It specifically addresses the complex serial-batching scheduling problem, where memory chips are tested on parallel machines under power constraints and chip deterioration effects. The processing time for each chip is significantly influenced by both the cumulative processing time of preceding chips and the power requirements for testing. We formulate this real-world optimization problem using a mixed integer nonlinear programming model. Exact solutions for small instances are obtained using a commercial solver. However, due to the model's complexity, we also develop heuristic solutions to efficiently handle larger instances. Based on the derivation of structural properties, we develop two tailored heuristic algorithms to determine the schedule for the final testing of memory chips. Additionally, we propose a refined Variable Neighborhood Search algorithm (VNS-H) that seamlessly integrates five local search strategies with two supplementary heuristic algorithms, dynamically alternating between them to ensure a balance between computational efficiency and the quality of the solutions obtained. Additionally, we establish a lower bound to validate the effectiveness of these solutions, particularly for large-scale instances. To validate the efficacy and robustness of our proposed metaheuristic algorithm, we conduct a rigorous comparison of the VNS-H algorithm with five other metaheuristic algorithms that have promising performance in various optimization problems. The results highlight the superior performance of the VNS-H algorithm. In small-scale instances, our VNS-H algorithm achieves an average makespan that is 5.52% lower compared to the original VNS. For large-scale instances, the VNS-H algorithm reduces the average makespan by 13.46% compared to VNS. Finally, we discuss the managerial implications of our findings, providing insights specifically tailored to semiconductor manufacturing enterprises based on the outcomes of this study.

Development of a total variation diminishing (TVD) sea ice transport scheme and its application in an ocean (SCHISM v5.11) and sea ice (Icepack v1.3.4) coupled model on unstructured grids

Abstract. As the demand for increased resolution and complexity in unstructured sea ice models is growing, higher demands are also placed on the sea ice transport scheme. In this study, we couple the Semi-implicit Cross-scale Hydro-science Integrated System Model (SCHISM, v5.11) with Icepack (v1.3.4), the column physics package of the Los Alamos sea ice model (CICE); a key step is to implement a total variation diminishing (TVD) transport scheme for the multi-class sea ice module in the coupled model. Compared with the second-order upwind scheme and the finite-element flux-corrected transport (FEM-FCT) scheme, the TVD transport scheme is overall superior when evaluated based on conservation, accuracy, efficiency (even with very high resolution), and strict monotonicity. Although it is slightly weaker than FEM-FCT in terms of accuracy alone, the TVD scheme still outperforms the other two schemes in comprehensive performance. The new coupled model outperforms the existing single-class ice model of SCHISM in the case of Lake Superior. For the Arctic Ocean case, it successfully reproduces the long-term changes in the sea ice extent, sea ice boundary, concentration observations from satellites, and thickness from in situ measurement.

Direct and Indirect Effects of Environmental and Socio-Economic Factors on COVID-19 in Africa Using Structural Equation Modeling

Understanding direct and indirect relationships of environmental, socio-economic and climate variables and the dynamics of epidemics is key to guiding targeted public health policy and interventions. This study investigates the direct and indirect effects of environmental and socio-economic factors on the COVID-19 dynamics in Africa (54 African countries from 2019 to 2021) using SEM approach. Specifically, the study aimed to: (i) assess the performance of two SEM estimation methods (Lisrel and PLS-SEM) in relationship to sample size (100, 200, 500, and 1000) and level of model complexity (No, two, and four indirect effects) and (ii) use the most performing SEM estimation method to examine direct and indirect effects of factors influencing the number of cases and deaths of COVID-19 in Africa. The results highlight a positive spatial correlation between factors such as temperature, humidity, age, the proportion of people aged over 65, and the COVID-19 incidence. Under the control of confounding factors, Lisrel turns out to be the most performing method, identifying climate, demographic and economic factors as the main determinants of COVID-19 dynamics. These factors have a direct and significant impact on the incidence of COVID-19. An indirect relationship was also observed between economic factors and the incidence of COVID-19 through air pollutants. The results highlight the importance of considering these factors in understanding the spread of the virus to avoid further disasters.

Research on Applying Deep Learning to Visual–Motor Integration Assessment Systems in Pediatric Rehabilitation Medicine

In pediatric rehabilitation medicine, manual assessment methods for visual–motor integration result in inconsistent scoring standards. To address these issues, incorporating artificial intelligence (AI) technology is a feasible approach that can reduce time and improve accuracy. Existing research on visual–motor integration scoring has proposed a framework based on convolutional neural networks (CNNs) for the Beery–Buktenica developmental test of visual–motor integration. However, as the number of training questions increases, the accuracy of this framework significantly decreases. This paper proposes a new architecture to reduce the number of features, channels, and overall model complexity. The architectureoptimizes input features by concatenating question numbers with answer features and selecting appropriate channel ratios and optimizes the output vector by designing the task as a multi-class classification. This paper also proposes a model named improved DenseNet. After experimentation, DenseNet201 was identified as the most suitable pre-trained model for this task and was used as the backbone architecture for improved DenseNet. Additionally, new fully connected layers were added for feature extraction and classification, allowing for specialized feature learning. The architecture can provide reasons for unscored results based on prediction results and decoding rules, offering directions for children’s training. The final experimental results show that the proposed new architecture improves the accuracy of scoring 6 question graphics by 12.8% and 12 question graphics by 20.14% compared to the most relevant literature. The accuracy of the proposed new architecture surpasses the model frameworks of the most relevant literature, demonstrating the effectiveness of this approach in improving scoring accuracy and stability.

YOLOSAR-Lite: a lightweight framework for real-time ship detection in SAR imagery

ABSTRACT Real-time ship detection using SAR images is crucial for maritime operations but remains challenging due to issues such as sidelobes, complex background interferences, and target defocusing. Studies have shown that deep learning can address these problems, current deep learning algorithms often suffer from achieving the necessary accuracy due to complex noise and background characteristics. Additionally, advanced algorithms excelling in feature extraction usually require significant computational resources. This paper proposes a novel deep learning approach based on YOLOv8 to improve accuracy while reducing model complexity and computational burden. The approach innovatively integrates a knowledge distillation module into the YOLOv8. We further enhance the model by replacing the original YOLOv8 backbone, neck, and head with lightweight alternatives: HGNetv2, SlimNeck, and a newly designed decoupled head. The enhanced model, YOLOSAR-Lite, demonstrated exceptional performance on both accuracy and model complexity metrics. Rigorous evaluation on the Official-SSDD dataset and the SAR-Ship-Dataset revealed recognition accuracies of 95.32% and 96.06%, respectively. Furthermore, YOLOSAR-Lite achieved significant reduction in complexity compared to the YOLOv8: 4.48G FLOPs (45.30% reduction), 2.05M parameters (31.89% reduction), and 3.91MB model size (31.76% reduction). These improvements demonstrate YOLOSAR-Lite achieves higher accuracy with minimal complexity. Consequently, YOLOSAR-Lite is a promising solution for real-time ship detection.

Dementia risk prediction using decision-focused content selection from medical notes

Several general-purpose language model (LM) architectures have been proposed with demonstrated improvement in text summarization and classification. Adapting these architectures to the medical domain requires additional considerations. For instance, the medical history of the patient is documented in the Electronic Health Record (EHR) which includes many medical notes drafted by healthcare providers. Direct processing of these notes may not be possible because the computational complexity of LMs imposes a limit on the length of input text. Therefore, previous applications resorted to content selection using truncation or summarization of the text. Unfortunately, these text processing techniques may lead to information loss, redundancy or irrelevance. In the present paper, a decision-focused content selection technique is proposed. The objective of this technique is to select a subset of sentences from the medical notes of a patient that are relevant to the target outcome over a predefined observation period. This decision-focused content selection methodology is then used to develop a dementia risk prediction model based on the Longformer LM architecture. The results show that the proposed framework delivers an AUC of 78.43 when the summary is restricted to 1024 tokens, outperforming previously proposed content selection techniques. This performance is notable given that the model estimates dementia risk with a one year prediction horizon, relies on an observation period of only one year and solely uses medical notes without other EHR data modalities. Moreover, the proposed techniques overcome the limitation of machine learning models that use a tabular representation of the text by preserving contextual content, enable feature engineering from raw text and circumvent the computational complexity of language models.

SPCANet: congested crowd counting via strip pooling combined attention network.

Crowd counting aims to estimate the number and distribution of the population in crowded places, which is an important research direction in object counting. It is widely used in public place management, crowd behavior analysis, and other scenarios, showing its robust practicality. In recent years, crowd-counting technology has been developing rapidly. However, in highly crowded and noisy scenes, the counting effect of most models is still seriously affected by the distortion of view angle, dense occlusion, and inconsistent crowd distribution. Perspective distortion causes crowds to appear in different sizes and shapes in the image, and dense occlusion and inconsistent crowd distributions result in parts of the crowd not being captured completely. This ultimately results in the imperfect capture of spatial information in the model. To solve such problems, we propose a strip pooling combined attention (SPCANet) network model based on normed-deformable convolution (NDConv). We model long-distance dependencies more efficiently by introducing strip pooling. In contrast to traditional square kernel pooling, strip pooling uses long and narrow kernels (1×N or N×1) to deal with dense crowds, mutual occlusion, and overlap. Efficient channel attention (ECA), a mechanism for learning channel attention using a local cross-channel interaction strategy, is also introduced in SPCANet. This module generates channel attention through a fast 1D convolution to reduce model complexity while improving performance as much as possible. Four mainstream datasets, Shanghai Tech Part A, Shanghai Tech Part B, UCF-QNRF, and UCF CC 50, were utilized in extensive experiments, and mean absolute error (MAE) exceeds the baseline, which is 60.9, 7.3, 90.8, and 161.1, validating the effectiveness of SPCANet. Meanwhile, mean squared error (MSE) decreases by 5.7% on average over the four datasets, and the robustness is greatly improved.

A Study on Network Anomaly Detection Using Fast Persistent Contrastive Divergence

As network technology evolves, cyberattacks are not only increasing in frequency but also becoming more sophisticated. To proactively detect and prevent these cyberattacks, researchers are developing intrusion detection systems (IDSs) leveraging machine learning and deep learning techniques. However, a significant challenge with these advanced models is the increased training time as model complexity grows, and the symmetry between performance and training time must be taken into account. To address this issue, this study proposes a fast-persistent-contrastive-divergence-based deep belief network (FPCD-DBN) that offers both high accuracy and rapid training times. This model combines the efficiency of contrastive divergence with the powerful feature extraction capabilities of deep belief networks. While traditional deep belief networks use a contrastive divergence (CD) algorithm, the FPCD algorithm improves the performance of the model by passing the results of each detection layer to the next layer. In addition, the mix of parameter updates using fast weights and continuous chains makes the model fast and accurate. The performance of the proposed FPCD-DBN model was evaluated on several benchmark datasets, including NSL-KDD, UNSW-NB15, and CIC-IDS-2017. As a result, the proposed method proved to be a viable solution as the model performed well with an accuracy of 89.4% and an F1 score of 89.7%. By achieving superior performance across multiple datasets, the approach shows great potential for enhancing network security and providing a robust defense against evolving cyber threats.

Lightweight omni-dimensional dynamic convolution with spatial-spectral self-attention network for hyperspectral image classification

ABSTRACT Omni-dimensional dynamic convolution can adjust spatial sizes, the amount of input/output channels, and kernel size based on the characteristics of the input, thus improving the capacity to extract characteristics through convolution, resulting in increased computation and model complexity. As a result, it is challenging to deploy these models on resource-limited airborne high-spectral testing systems. This paper proposes a lightweight omni-dimensional dynamic convolution network as a possible solution to this issue. In this network, the LS2CM module is used for spatial-spectral information extraction, and the pyramidal residual network is the basic architecture for extracting features from shallow to deep layers, and lightweight omni-dimensional dynamic convolution is adopted in the convolution module. Overcoming the pyramidal structure can only extract local information, we proposed a new simple spatial-spectral self-attention mechanism to extract global information from hyperspectral. Besides, the Huber function as an activation function. The IP, WHU-Hi-LongKou, SA, and PU datasets were used to verify the accuracy of the suggested network model. The results show that the proposed lightweight network maintains an appropriate equilibrium among model complexity as well as accuracy.

Prune and Quantize Semantic Segmentation Network for Aerial Objects Recognition

Semantic segmentation of aerial and satellite images is crucial for applications in environmental management, urban planning, and traffic safety. While deep learning techniques with convolutional neural networks (CNNs) and attention mechanisms have achieved superior accuracy compared to traditional methods, they often struggle with model complexity and resource constraints. This paper introduces two novel techniques - pruning and quantization - to enhance both the performance and efficiency of semantic segmentation models for remote sensing images (RSIs). Pruning reduces model complexity by eliminating less significant weights, while quantization decreases memory usage by converting weights into a more compact format. We applied these techniques to the DeepLabV3+ model with ResNet18 and ResNet50 backbones and assessed their performance across multiple RSI datasets. Our results show that pruning and quantization effectively balance accuracy and computational efficiency, achieving a mean IoU of 81.24% with a memory footprint of 135.19 MB for pruning, and 81.04% mean IoU with a memory footprint of 33.79 MB for quantization on the ISPRS Vaihingen dataset. These methods offer a viable solution for deploying semantic segmentation models on resource-constrained hardware.