semantic segmentation
در نشریات گروه برق-
Protecting privacy in street view imagery is a critical challenge in urban analytics, requiring comprehensive and scalable solutions beyond localized obfuscation techniques such as face or license plate blurring. To address this, we propose a novel framework that automatically detects and removes sensitive objects, such as pedestrians and vehicles, ensuring robust privacy preservation while maintaining the visual integrity of the images. Our approach integrates semantic segmentation with 2D priors and multimodal data from cameras and LiDAR to achieve precise object detection in complex urban scenes. Detected regions are seamlessly filled using a large-mask inpainting technique based on fast Fourier convolutions (FFC), enabling efficient generalization to high-resolution imagery. Evaluated on the SemanticKITTI dataset, our method achieves a mean Intersection over Union (mIoU) of 64.9%, surpassing state-of-the-art benchmarks. Despite its reliance on accurate sensor calibration and multimodal data availability, the proposed framework offers a scalable solution for privacy-sensitive applications such as urban mapping, and virtual tourism, delivering high-quality anonymized imagery with minimal artifacts.
Keywords: Privacy Protection, Street View Imagery, Large Mask Inpainting, Semantic Segmentation, Multi-Modality, Lidar -
Semantic segmentation is a critical task in computer vision, focused on extracting and analyzing detailed visual information. Traditional artificial neural networks (ANNs) have made significant strides in this area, but spiking neural networks (SNNs) are gaining attention for their energy efficiency and biologically inspired time-based processing. However, existing SNN-based methods for semantic segmentation face challenges in achieving high accuracy due to limitations such as quantization errors and suboptimal membrane potential distribution. This research introduces a novel spiking approach based on Spiking-DeepLab, incorporating a Regularized Membrane Potential Loss (RMP-Loss) to address these challenges. Built upon the DeepLabv3 architecture, the proposed model leverages RMP-Loss to enhance segmentation accuracy by optimizing the membrane potential distribution in SNNs. By optimizing the storage of membrane potentials, where values are stored only at the final time step, the model significantly reduces memory usage and processing time. This enhancement not only improves the computational efficiency but also boosts the accuracy of semantic segmentation, enabling more accurate temporal analysis of network behavior. The proposed model also demonstrates better robustness against noise, maintaining its accuracy under varying levels of Gaussian noise, which is common in real-world scenarios. The proposed approach demonstrates competitive performance on standard datasets, showcasing its potential for energy-efficient image processing applications.
Keywords: Supervised Learning, Image Processing, Semantic Segmentation, Spiking Neural Networks, RMP-Loss -
In recent years, Convolutional Neural Networks (CNNs) have made significant strides in the field of segmentation, particularly in semantic segmentation where both accuracy and efficiency are crucial. However, despite their high accuracy, these deep networks are not practical for real-time use due to their low inference speed. This issue has prompted researchers to explore various techniques to improve the efficiency of CNNs. One such technique is knowledge distillation, which involves transferring knowledge from a larger, cumbersome (teacher) model to a smaller, more compact (student) model. This paper proposes a simple yet efficient approach to address the issue of low inference speed in CNNs using knowledge distillation. The proposed method involves distilling knowledge from the feature maps of the teacher model to guide the learning of the student model. The approach uses a straightforward technique known as pixel-wise distillation to transfer the feature maps of the last convolution layer of the teacher model to the student model. Additionally, a pair-wise distillation technique is used to transfer pair-wise similarities of the intermediate layers. To validate the effectiveness of the proposed method, extensive experiments were conducted on the PascalVoc 2012 dataset using a state-of-the-art DeepLabV3+ segmentation network with different backbone architectures. The results showed that the proposed method achieved a balanced mean Intersection over Union (mIoU) and training time.Keywords: Computer Vision, Convolutional Neural Networks, Deep learning, Semantic Segmentation, Knowledge Distillation, Deep Neural Networks
-
بخش بندی معنایی عمیق تصاویر به عنوان راه حلی یکپارچه در آنالیز تصاویر مبتنی بر طبقه بندی تک تک پیکسل های تصویر بوده و بخصوص در کاربرد هایی مانند شناسایی نشت نفت در آب های آزاد که در آن مرز اشیا و نواحی به طور مشخص قابل تفکیک نیستند، مورد توجه قرار می گیرد. به منظور کنترل هرچه بیشتر آلودگی و مخاطرات زیست محیطی ناشی از نشت نفت، ارایه روش هایی با دقت بیشتر از اهمیت ویژه ای برخوردار است. تصاویر رادار روزنه مصنوعی دراین زمینه بسیار پرکاربرد بوده و با چالش هایی از جمله نویز اسپکل و نیز تشخیص نواحی لکه نفتی و شبه لکه نفتی مواجه هستند. بکارگیری روش های نوین یادگیری عمیق می تواند در کاهش دخالت سلیقه انسانی در تصمیم گیری کمک کند. در این مقاله از روش مخلوط کردن کانال های ویژگی در شبکه های کانولوشنی عمیق، بلوک های آتروس و بخش های رمزگشایی استفاده شده است که علاوه بر کاهش پیچیدگی های محاسباتی، نتایج بخش بندی لکه های نفتی به مراتب بهتر از سایر روش ها می دهد. معماری شبکه ارایه شده مبتنی بر معماری vgg16 می باشد. دقت کلی، صحت، همپوشانی بر واحد، IoU وزن دار و امتیاز BF به عنوان پارامترهای ارزیابی در نظر گرفته شده اند. در روش ارایه شده، دقت بخش بندی لکه های نفتی و شبه لکه های نفتی به ترتیب به میزان 8/7% و 3/7% نسبت به روش های پیشین بهبود یافته است.
کلید واژگان: مخلوط کردن کانال، شافل نت، یادگیری عمیق، بخش بندی معنایی، شناسایی لکه های نفتی، تصاویر رادار روزنه مصنوعیJournal of Iranian Association of Electrical and Electronics Engineers, Volume:19 Issue: 3, 2022, PP 131 -144Deep Semantic segmentation of images as an integrated solution for image analysis is based on the classification of individual image pixels, especially in applications such as oil spill detection in the marine areas which have no clear boundaries. To curb pollution and environmental hazards caused by oil spills, it is important to provide more accurate algorithms. Synthetic aperture radar images are widely used in the oil spill detection field. In these images, there are challenges such as speckle noise as well as distinguishing between oil spills and lookalike areas. The application of new machine learning methods can help reduce the involvement of human taste in decision-making. In this paper, the feature channel shuffling method on CNN networks, atrous block, and decoder parts are used and the computational complexity is drastically reduced and also provides much better oil spill segmentation results than other methods. The proposed network architecture is based on the Vgg16 architecture. The overall accuracy, accuracy, intersection over :union:, weighted IoU, and BF score is used as evaluation parameters. In the proposed method, the accuracy of detecting the oil spills and look-alikes was improved by 7.8% and 7.3%, respectively, compared to the previous simulated methods.
Keywords: Channel shuffle, ShuffleNet, deep learning, semantic segmentation, oil spill detection, synthetic aperture radar images -
This paper presents a semantic method for aerial image segmentation. Multi-class aerial images are often featured with large intra-class variations and inter-class similarities. Furthermore, shadows, reflections and changes in viewpoint, high and varying altitude and variability of natural scene pose serious problems for simultaneous segmentation. The main purpose of segmentation of aerial images is to make subsequent recognition phase straightforward. Present algorithm combines two challenging tasks of segmentation and classification in a manner that no extra recognition phase is needed. This algorithm is supposed to be part of a system which will be developed to automatically locate the appropriate site for Unmanned Aerial Vehicle (UAV) landing. With this perspective, we focused on segregating natural and man-made areas in aerial images. We compared different classifiers and explored the best set of features for this task in an experimental manner. In addition, a certainty based method has been used for integrating color and texture descriptors in a more efficient way. The experimental results over a dataset comprised of 25 high-resolution images show the overall binary segmentation accuracy rate of 91.34%.Keywords: Aerial Images, Semantic Segmentation, Classification, Local Binary Patterns, Feature Fusion, Artificial Neural Network, Support Vector Machine, Random Forest
- نتایج بر اساس تاریخ انتشار مرتب شدهاند.
- کلیدواژه مورد نظر شما تنها در فیلد کلیدواژگان مقالات جستجو شدهاست. به منظور حذف نتایج غیر مرتبط، جستجو تنها در مقالات مجلاتی انجام شده که با مجله ماخذ هم موضوع هستند.
- در صورتی که میخواهید جستجو را در همه موضوعات و با شرایط دیگر تکرار کنید به صفحه جستجوی پیشرفته مجلات مراجعه کنید.