Published: Aug. 23, 2024
Language: Английский
The international archives of the photogrammetry, remote sensing and spatial information sciences/International archives of the photogrammetry, remote sensing and spatial information sciences, Journal Year: 2023, Volume and Issue: XLVIII-1/W2-2023, P. 1749 - 1755
Published: Dec. 14, 2023
Abstract. This study aims to enhance the quality of detecting burned areas in satellite imagery using deep learning by optimizing training dataset volume through application various augmentation methods. The analyzes impact image flipping, rotation, and noise addition on overall accuracy for different classes a forest: fire, burned, smoke background. Results demonstrate that while single techniques such as flipping rotation alone did not result significant improvements, combined approach resulted an enhancement classification accuracy. Moreover, shows augmenting use multiple methods concurrently, resulting fivefold increase input data, also enhanced recognition highlights need further research developing more efficient CNN models experimenting with additional improve area detection, which would benefit environmental protection emergency response services.
Language: Английский
Citations
2
Indonesian Journal of Electrical Engineering and Computer Science, Journal Year: 2024, Volume and Issue: 34(2), P. 1344 - 1344
Published: March 23, 2024
This paper addresses the classification problem in machine learning, focusing on predicting class labels for datasets with continuous features. Recognizing critical role of discretization enhancing performance, study integrates equal width binning (EWB) two optimization algorithms: bat algorithm (BA), referred to as EB, and whale (WOA), denoted EW. The primary objective is determine optimal technique relevant labels. emphasizes significance data preprocessing, offering a comprehensive approach that combines techniques algorithms. An investigative was undertaken assess efficiency EB EW by evaluating their performance using Naive Bayes K-nearest neighbor algorithms four sourced from UCI datasets. According experimental findings, suggested has major effect accuracy, recall, F-measure classification. outperforms other existing approaches all
Language: Английский
Citations
0
Published: Feb. 23, 2024
The electricity consumption of Chinese users is constantly increasing with the development economy. With continuous enrichment residents' lives and diversification activity forms, consumption, characteristics types among different present rich, diverse characteristics. In order to achieve effective for users, ensure reliability safety user power supply, energy conservation, emission reduction sustainable development, ensuring a safer more reliable supply has become basic requirement responsibility sector today. This paper discussed demand forecasting method important based on big data neural network. First, it briefly introduced network construction model data. Finally, comparative experimental analysis verified that accuracy LSTM (Long Short Term Memory networks) was higher than (the average value MAPE decreased by 0.689%).
Language: Английский
Citations
0
Published: Feb. 23, 2024
In this paper, we utilize the mechanism of simulating human nervous system in deep learning for movie data. Feature extraction is performed automatically and unstructured data processed to improve model recommendation effect. Through loss function output layer, iterative optimization carried out with help back propagation algorithm, which passes prediction error from layer input reverse, updates network parameters each turn. The analysis found that accuracy algorithm based on increases faster, point after 160 rounds leveled off, compared optimal improved by 1.4%. can better meet user's personalized needs recommend best meets requirements.
Language: Английский
Citations
0
Published: Feb. 23, 2024
Amidst the rapid evolution of large data technology, need to swiftly and accurately filter information from datasets has become paramount. This paper introduces a novel machine learning model denominated as 'WOA-BP', which synergistically aligns Whale Optimization Algorithm (WOA) with Backpropagation (BP) Neural Network. The newly proposed leverages WOA's convergence robust global search capabilities optimize weight biases BP neural network. To assess its performance, WOA-BP is benchmarked against five industrial models using publicly available datasets. Notably, algorithm achieves an outstanding regression prediction score (R2 value) 0.954 along impressive classification accuracy 95.56%, both are highest among all evaluated. It worth highlighting that WOA-optimized (e.g., WOA-Decision Trees WOA-Bayesian) consistently outperform their counterparts conventional Decision Bayesian models). These findings highlight promise WOA for future applications in enhancing addressing complex optimization challenges.
Language: Английский
Citations
0
Published: Feb. 23, 2024
Microwave antenna receivers face susceptibility to electrostatic discharge interference. This study explores surface corona and spark effects on the receiver, analyzing their time-frequency domain characteristics. Additionally, it investigates coupling effect of within shielding chassis microwave receiver system through full-wave simulation. The research assesses interference voltage internal circuit during discharge, evaluating its potential impact receiver's normal operation. Results indicate under IEC6100-4-2 (EMC) standard, causing read-write program disruption with critical values 17 5. offers insights for designing systems anti-electrostatic equipment.
Language: Английский
Citations
0
Published: Feb. 23, 2024
In mobile communication systems, channel attack and defense have always been one of the most important highly concerned topics in field wireless interference technology research application. current environment, all types equipment face threats to varying degrees. This article is dedicated analyzing discussing how use side channels for signal transmission data collection, also proposes various methods based on error recovery technology, multiplexers, etc. deal with possible loss damage problems terminals. By evaluating rate, security other factors sending receiving ends, source can be determined whether it exists. Finally, through system testing experiments, was verified that trace number range original method 156-176, while advanced encryption standard algorithm 134-153; training accuracy within 64%-79%, Advanced Encryption Standard 80%-91%; isolation capability 0.63-0.69, 0.84-0.93. Comprehensive experimental results show performs very well aspects.
Language: Английский
Citations
0
Published: Feb. 23, 2024
The fusion of content-based recommendation and collaborative filtering is an important research direction in algorithms. However, due to the scalability processing data, final results are often inaccurate. This article proposes a hybrid tourism algorithm based on central clustering parameters. Firstly, improve K-means algorithm, secondly, process, finally, perform weighted two experimental showed that issue was improved, accuracy also improved.
Language: Английский
Citations
0
Published: Feb. 23, 2024
The digital economy is becoming the main force driving development of world economy, and its sustainable cannot be achieved without sound governance. Establishing an objective scientific evaluation index system great significance for accurately understanding level promoting global On basis analyzing connotation it was summarized as four major characteristics: "economies scale", "technology driven", "ecological development", "dynamic management". Guided by characteristics theories such innovation systems, ecosystems, new institutional economics, a systematic framework risk assessment established based on principles. Starting from phenomenon, causes, era value risks, this article analyzed necessity governance starting point prevention in China, provided countermeasures legal prevention. It also explored construction warning combining BPNN (Back propagation neural network) GA (Genetic Algorithm) algorithm. Finally, experiment proved that accuracy predicting "good economy" "severe economic warning" over 94%.
Language: Английский
Citations
0
Published: Feb. 23, 2024
To effectively address information overload and knowledge redundancy, search recommendation have emerged as an effective way for users to improve retrieval efficiency in massive amounts of information. Traditional ecommerce systems cannot recommend agricultural products based on their unique attributes. The graph stores rich semantic nodes edges, enabling deep mining analysis entities relationships. This article adopts a top-down approach construct e-commerce, integrates attention mechanisms model, updates the embedding its own by recursively propagating neighboring graph. Additionally, are used distinguish importance more accurately capture user interests. research results provide solution constructing product system. By typing keywords engines, can quickly find related needs, providing accurate personalized services.
Language: Английский
Citations
0