An intelligent water drop algorithm with deep learning driven vehicle detection and classification DOI Creative Commons
Thavavel Vaiyapuri,

M. Sivakumar,

S. Shridevi

et al.

AIMS Mathematics, Journal Year: 2024, Volume and Issue: 9(5), P. 11352 - 11371

Published: Jan. 1, 2024

<abstract> <p>Vehicle detection in Remote Sensing Images (RSI) is a specific application of object recognition like satellite or aerial imagery. This highly beneficial different fields defense, traffic monitoring, and urban planning. However, complex particulars about the vehicles surrounding background, delivered by RSIs, need sophisticated investigation techniques depending on large data models. crucial though amount reliable labelled training datasets still constraint. The challenges involved vehicle from RSIs include variations orientations, appearances, sizes due to dissimilar imaging conditions, weather, terrain. Both architecture hyperparameters Deep Learning (DL) algorithm must be tailored features RS nature tasks. Therefore, current study proposes Intelligent Water Drop Algorithm with Learning-Driven Vehicle Detection Classification (IWDADL-VDC) methodology applied upon Images. IWDADL-VDC technique exploits hyperparameter-tuned DL model for both classification vehicles. In order accomplish this, follows two major stages, namely classification. For process, method uses improved YOLO-v7 model. After are detected, next stage performed help Long Short-Term Memory (DLSTM) approach. enhance outcomes DLSTM model, IWDA-based hyperparameter tuning process has been employed this study. experimental validation was conducted using benchmark dataset results attained were promising over other recent approaches.</p> </abstract>

Language: Английский

Effective segmentation of land-use and land-cover from hyperspectral remote sensing image DOI
Vijaykumar P. Yele, Sujata Alegavi, R. R. Sedamkar

et al.

International Journal of Information Technology, Journal Year: 2024, Volume and Issue: 16(4), P. 2395 - 2412

Published: Jan. 26, 2024

Language: Английский

Citations

4
THE METHOD FOR OBJECTS DETECTION ON SATELLITE IMAGERY BASED ON THE FIREFLY ALGORITHM DOI Creative Commons
Hennadii Khudov, Irina Khizhnyak, Sergey Glukhov

et al.

Advanced Information Systems, Journal Year: 2024, Volume and Issue: 8(1), P. 5 - 11

Published: Feb. 26, 2024

The subject matter of the article is method for detecting objects on satellite imagery based firefly algorithm. goal to develop a tasks are: analysis existing methods interest imagery, development practical verification algorithm, and quantitative assessment quality proposed method. used digital image processing, data clustering, mathematical apparatus matrix theory, swarm intelligence, modeling, optimization analytical empirical comparison. following results are obtained. advantages disadvantages main approaches processing purpose them determined. general principle operation algorithm considered. It presents flowchart in one color channel. values input parameters were selected experimentally. Experimental studies conducted real errors first second kind processed using particle calculated. Conclusions. Analysis calculated showed that compared algorithm: reduces error by about 11% 9%. directions further research study problem selecting

Language: Английский

Citations

4
Flood Mapping and Damage Assessment using Ensemble Model Approach DOI
Vrushabh Patil, Yadnyadeep Khadke, Amit Joshi

et al.

Sensing and Imaging, Journal Year: 2024, Volume and Issue: 25(1)

Published: March 4, 2024

Language: Английский

Citations

4
Hybridizing Deep Neural Networks and Machine Learning Models for Aerial Satellite Forest Image Segmentation DOI Creative Commons
Clopas Kwenda, Mandlenkosi Gwetu, Jean Vincent Fonou-Dombeu

et al.

Journal of Imaging, Journal Year: 2024, Volume and Issue: 10(6), P. 132 - 132

Published: May 29, 2024

Forests play a pivotal role in mitigating climate change as well contributing to the socio-economic activities of many countries. Therefore, it is paramount importance monitor forest cover. Traditional machine learning classifiers for segmenting images lack ability extract features such spatial relationship between pixels and texture, resulting subpar segmentation results when used alone. To address this limitation, study proposed novel hybrid approach that combines deep neural networks algorithms segment an aerial satellite image into non-forest regions. Aerial were first extracted by two network models, namely, VGG16 ResNet50. The are subsequently five including Random Forest (RF), Linear Support Vector Machines (LSVM), k-nearest neighbor (kNN), Discriminant Analysis (LDA), Gaussian Naive Bayes (GNB) perform final segmentation. obtained from globe challenge dataset. performance model was evaluated using metrics Accuracy, Jaccard score index, Root Mean Square Error (RMSE). experimental revealed RF achieved best with accuracy, score, RMSE 94%, 0.913 0.245, respectively; followed LSVM 89%, 0.876, 0.332, respectively. LDA took third position 88%, 0.834, 0.351, respectively, GNB 0.837, 0.353, kNN occupied last 83%, 0.790, 0.408, also has significantly improved RF, LSVM, LDA, compared their Furthermore, showed outperformed other models related studies, thereby, attesting its superior capability.

Language: Английский

Citations

4
A novel W13 deep CNN structure for improved semantic segmentation of multiple objects in remote sensing imagery DOI Creative Commons
Khaled Mohammed Elgamily, Mohamed A. Mohamed, Ahmed Aboutaleb

et al.

Neural Computing and Applications, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 3, 2025

Abstract This paper proposes a novel convolutional neural network (CNN) architecture designed for semantic segmentation in remote sensing images. The proposed W13 Net model addresses the inherent challenges of tasks through carefully crafted architecture, combining strengths multistage encoding–decoding, skip connections, combined weighted output, and concatenation techniques. Compared with different models, suggested performs better. A comprehensive analysis models has been carried out, resulting an extensive comparison between five existing state-of-the-art architectures. Utilizing two standardized datasets, Dense Labeling Remote Sensing Dataset Termed (DLRSD), Mohammad Bin Rashid Space Center (MBRSC) Dubai Aerial Imagery Dataset, evaluation entails training, testing, validation across classes. demonstrates adaptability, generalization capabilities, superior results key metrics, all while displaying robustness variety datasets. number including accuracy, precision, recall, F1 score, IOU, were used to evaluate system’s performance. According experimental results, obtained accuracy 87.8%, precision 0.88, recall score IOU 0.74. showed significant improvement increase up 18%, when compared other recent taking into consideration model’s comparatively low parameter (2.2 million) models.

Language: Английский

Citations

0
Farmland Segmentation in Landsat 8 Satellite Images Using Deep Learning and Conditional Generative Adversarial Networks DOI Creative Commons
Shruti Nair, Sara Sharifzadeh, Vasile Palade

et al.

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(5), P. 823 - 823

Published: Feb. 27, 2024

Leveraging mid-resolution satellite images such as Landsat 8 for accurate farmland segmentation and land change monitoring is crucial agricultural management, yet hindered by the scarcity of labelled data training supervised deep learning pipelines. The particular focus this study on addressing images. This paper introduces several contributions, including a systematic image augmentation approach that aims to maintain population consistency during model training, thus mitigating performance degradation. To alleviate labour-intensive task pixel-wise labelling, we present novel application modified conditional generative adversarial network (CGAN) generate artificial corresponding farm labels. Additionally, scrutinize role spectral bands in compare two prominent semantic models, U-Net DeepLabV3+, with diverse backbone structures. Our empirical findings demonstrate augmenting dataset up 22.85% samples significantly enhances performance. Notably, model, employing standard convolution, outperforms DeepLabV3+ models atrous achieving accuracy 86.92% test data.

Language: Английский

Citations

3
Interannual Monitoring of Cropland in South China from 1991 to 2020 Based on the Combination of Deep Learning and the LandTrendr Algorithm DOI Creative Commons
Yue Qu, Boyu Zhang, Han Xu

et al.

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(6), P. 949 - 949

Published: March 8, 2024

Timely and accurate acquisition of spatial distribution changes in cropland is significant importance for food security ecological preservation. Most studies that monitor long-term tend to overlook the rationality process evolution, there are conflicts between interannual data, so they cannot be used analyze land use change. This study focuses on annual identification results cropland, considering evolution short-term variations influenced by natural environmental human activities. An approach monitoring based long time series deep learning also proposed. We acquired imagery related cropland’s vegetation lush period (VLP) differential (VDP) from Landsat images Google Earth Engine (GEE) platform ResUNet-a structural model training. Finally, a long-time-series correction algorithm LandTrendr introduced, maps Guangdong Province 1991 2020 were generated. Evaluating every five years, we found an overall accuracy 0.91–0.93 kappa coefficient 0.80–0.83. Our demonstrate good consistency with agricultural statistical data. Over past 30 total area has undergone three phases: decrease, stabilization. Significant regional have been observed. can applied southern regions China, providing valuable data support further implementation protection.

Language: Английский

Citations

3
Detection of color phenotype in strawberry germplasm resources based on field robot and semantic segmentation DOI

Ningyuan Yang,

Zhenyu Huang, Yong He

et al.

Computers and Electronics in Agriculture, Journal Year: 2024, Volume and Issue: 226, P. 109464 - 109464

Published: Sept. 23, 2024

Language: Английский

Citations

2
On the Robustness and Generalization Ability of Building Footprint Extraction on the Example of SegNet and Mask R-CNN DOI Creative Commons
Muntaha Sakeena,

Eric Stumpe,

Miroslav Despotović

et al.

Remote Sensing, Journal Year: 2023, Volume and Issue: 15(8), P. 2135 - 2135

Published: April 18, 2023

Building footprint (BFP) extraction focuses on the precise pixel-wise segmentation of buildings from aerial photographs such as satellite images. BFP is an essential task in remote sensing and represents foundation for many higher-level analysis tasks, disaster management, monitoring city development, etc. challenging because can have different sizes, shapes, appearances both same region regions world. In addition, effects, occlusions, shadows, bad lighting, to also be considered compensated. A rich body work has been presented literature, promising research results reported benchmarking datasets. Despite comprehensive performed, it still unclear how robust generalizable state-of-the-art methods are regions, cities, settlement structures, densities. The purpose this study close gap by investigating questions practical applicability extraction. particular, we evaluate robustness generalizability well their transfer learning capabilities. Therefore, investigate detail two most popular deep architectures (i.e., SegNet, encoder–decoder-based architecture Mask R-CNN, object detection architecture) them with respect aspects a proprietary high-resolution image dataset publicly available Results show that networks generalize new data, across cities continents. They benefit increased training especially when data distribution (data source) or comparable resolution. Transfer source recording parameters not always beneficial.

Language: Английский

Citations

4
Hızlandırılmış Makine Öğrenmesi Algoritmaları ile Tarım Parseli Tabanlı Ürün Desen Sınıflandırması DOI
Fatih Fehmi Şimşek

Yüzüncü Yıl Üniversitesi Fen Bilimleri Enstitüsü Dergisi, Journal Year: 2024, Volume and Issue: 29(1), P. 314 - 330

Published: April 18, 2024

Gelişen teknoloji sayesinde, uydu görüntüleri ve uzaktan algılama çalışmaları, tarım alanında öncü çalışmalar arasında yer almaktadır. Tarımsal ürün desen tespitinde en yaygın kullanılan yöntemlerin başında ise teknolojisi gelmektedir. Uydu ile oluşturulan haritaları, Tarım Orman Bakanlığı tarafından destekleme ödemelerinde altlık olarak aktif bir şekilde kullanılmaktadır. Bu çalışmada, çalışma alanı Eskişehir İli, Seyitgazi Sivrihisar İlçe sınırları içerisinde kalan alan seçilmiş, çok zamanlı Sentinel-2 hızlandırılmış makine öğrenme algoritmaları (GBM, XGBoost, LightGBM, CatBoost) kullanılarak obje tabanlı (tarım parseli) sınıflandırma çalışması yapılmış sonuçlar karşılaştırılmıştır. Yapılan sonucunda her algoritma %90 üzerinde genel doğruluk değerine ulaşılmıştır (GBM- %90.3, XGBoost-%91.1, LightGBM-%93.9, CatBoost-%93.5). Sınıflandırma çalışmasında parselleri kullanılmıştır. Çalışma parsel ekim yapılan sınırların bazı parsellerde farklılık gösterdiği, ayrıca parseli birden fazla farklı ürüne ait tarımsal üretim yapıldığı gözlemlenmiştir. parsellerinin kullanılması için sınırlarının sınırlarına göre düzenlenmesi/bölünmesi gerektiği sonucuna ulaşılmıştır. küçük ölçekli orta alanlarda uygulanan yöntem kullanılabilir olduğu, geniş alternatif yöntemin geliştirilmesi varılmıştır.

Citations

1