A Novel Deep Learning Architecture for Agriculture Land Cover and Land Use Classification from Remote Sensing Images Based on Network-Level Fusion of Self-Attention Architecture

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Journal Year: 2024, Volume and Issue: 17, P. 6338 - 6353

Published: Jan. 1, 2024

AI-driven precision agriculture applications can benefit from the large data source that remote sensing provides, as it gather agricultural monitoring at various scales throughout year. Numerous advantages for sustainable applications, including yield prediction, crop monitoring, and climate change adaptation, be obtained artificial intelligence. In this work, we proposed a fully automated Optimized Self-Attention Fused Convolutional Neural Network (CNN) architecture land use cover classification using (RS) data. A new contrast enhancement equation has been utilized in augmentation. After that, fused CNN was proposed. The initially consists of two custom models named IBNR-65 Densenet-64. Both have designed based on inverted bottleneck residual mechanism Dense Blocks. both were depth-wise concatenation append layer deep features extraction. trained model performed neural network (NN) classifiers. results NN classifiers are insufficient; therefore, implemented Bayesian Optimization fine-tuned hyperparameters NN. addition, Quantum Hippopotamus Algorithm best feature selection. selected finally classified improved accuracy 98.20, 89.50, 91.70%, highest rate is 98.23, recall f1-score 98.21 respectively, SIRI-WHU, EuroSAT, NWPU datasets. Moreover, detailed ablation study conducted, performance compared with SOTA. shows accuracy, sensitivity, computational time performance.

Language: Английский

---

Improving Safety and Efficiency of Industrial Vehicles by Bio‐Inspired Algorithms

Expert Systems, Journal Year: 2025, Volume and Issue: 42(3)

Published: Jan. 22, 2025

ABSTRACT In the context of industrial automation, optimising automated guided vehicle (AGV) trajectories is crucial for enhancing operational efficiency and safety. They must travel in crowded work areas cross narrow corridors with strict safety time requirements. Bio‐inspired optimization algorithms have emerged as a promising approach to deal complex scenarios. Thus, this paper explores ability three novel bio‐inspired algorithms: Bat Algorithm (BA), Whale Optimization (WOA) Gazelle (GOA); optimise AGV path planning environments. To do it, new strategy described: trajectory based on clothoid curves specialised piece‐wise fitness function which prioritises designed. Simulation experiments were conducted across different occupancy maps evaluate performance each algorithm. WOA demonstrates faster providing suitable solutions 4 times than GOA. Meanwhile, GOA gives better metrics but demands more computational time. The study highlights potential approaches optimisation suggests avenues future research, including hybrid algorithm development.

Language: Английский

---

DLAN: A Dual Attention Network for Effective Land Cover Classification in Remote Sensing

Knowledge-Based Systems, Journal Year: 2025, Volume and Issue: unknown, P. 113620 - 113620

Published: April 1, 2025

Language: Английский

---

An Integrated Parallel Inner Deep Learning Models Information Fusion With Bayesian Optimization for Land Scene Classification in Satellite Images

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Journal Year: 2023, Volume and Issue: 16, P. 9888 - 9903

Published: Jan. 1, 2023

Classification of remote scenes in satellite imagery has many applications, such as surveillance, earth observation, etc. Classifying high-resolution sensing images machine learning is a big challenge nowadays. Several automated techniques based on and deep have been introduced the literature; however, these fail to perform for complex texture images, backgrounds, small objects. In this work, we proposed new technique inner fusion two models feature selection. A network designed at initial phase inner-level networks combined weights. After that, hyperparameters initialized Bayesian optimization (BO). Usually, through manual approach, but that not an efficient way model trained extracted features from deeper layer. last step, Poor-Rich controlled entropy-based selection developed best The selected are finally classified using classifiers. We performed experimental process architecture three publically available datasets: AID, UC-Merceds, WHU-RS19. On datasets, obtained accuracy 96.3%, 95.6%, 97.8%, respectively. Comparison conducted with state-of-the-art shows improved accuracy.

Language: Английский

---

A Deep Learning Framework with an Intermediate Layer Using the Swarm Intelligence Optimizer for Diagnosing Oral Squamous Cell Carcinoma

Diagnostics, Journal Year: 2023, Volume and Issue: 13(22), P. 3461 - 3461

Published: Nov. 16, 2023

One of the most prevalent cancers is oral squamous cell carcinoma, and preventing mortality from this disease primarily depends on early detection. Clinicians will greatly benefit automated diagnostic techniques that analyze a patient's histopathology images to identify abnormal lesions. A deep learning framework was designed with an intermediate layer between feature extraction layers classification for classifying histopathological into two categories, namely, normal carcinoma. The constructed using proposed swarm intelligence technique called Modified Gorilla Troops Optimizer. While there are many optimization algorithms used in literature selection, weight updating, optimal parameter identification models, work focuses as convert extracted features better suited classification. Three datasets comprising 2784 3632 carcinoma subjects considered work. popular CNN architectures, InceptionV2, MobileNetV3, EfficientNetB3, investigated layers. Two fully connected Neural Network layers, batch normalization, dropout With best accuracy 0.89 among examined MobileNetV3 exhibits good performance. This increased 0.95 when suggested Optimizer intermediary layer.

Language: Английский

---

Contextual based hybrid classification with FCM to handle mixed pixels and edge preservation

International Journal of Information Technology, Journal Year: 2024, Volume and Issue: 16(6), P. 3537 - 3547

Published: June 19, 2024

Language: Английский

---

A hybrid smell agent symbiosis organism search algorithm for optimal control of microgrid operations

PLoS ONE, Journal Year: 2023, Volume and Issue: 18(6), P. e0286695 - e0286695

Published: June 7, 2023

This paper presents a hybrid Smell Agent Symbiosis Organism Search Algorithm (SASOS) for optimal control of autonomous microgrids. In microgrid operation, single optimization algorithm often lacks the required balance between accuracy and speed to power system parameters such as frequency voltage effectively. The reduces imbalance exploitation exploration increases effectiveness in To achieve this, various energy resource models were coordinated into model generation distribution loads. problem was formulated based on network flow discrete-time sampling constrained parameters. development SASOS comprises components Symbiotic (SOS) Optimization (SAO) codified an loop. Twenty-four standard test function benchmarks used evaluate performance developed. experimental analysis revealed that obtained 58.82% Desired Convergence Goal (DCG) 17 benchmark functions. implemented Microgrid Central Controller (MCC) benchmarked alongside SOS SAO strategies. MATLAB/Simulink simulation results load disturbance rejection showed viability with improved reduction Total Harmonic Distortion (THD) 19.76%, compared SOS, SAO, MCC methods have THD 15.60%, 12.74%, 6.04%, respectively, over benchmark. Based obtained, it can be concluded demonstrates superior other methods. finding suggests is promising solution enhancing It also shown apply sectors engineering optimization.

Language: Английский

---

Sign Language Recognition Using Improved Seagull Optimization Algorithm with Deep Learning Model

Published: Aug. 28, 2024

Language: Английский

---

Authentic Signature Verification Using Deep Learning Embedding With Triplet Loss Optimization And Machine Learning Classification

International Journal of Computing and Digital Systems, Journal Year: 2024, Volume and Issue: 15(1), P. 265 - 277

Published: April 23, 2024

Various document types (financial, commercial, judicial) necessitate signatures for authentication.With the advancements of technology and increasing number documents, traditional signature verification methods encounter challenges in facing tasks related to verifying images, such as verification.This idea is further reinforced by growing migration transactions digital platforms.To that end, fields Machine learning (ML) Deep Learning (DL) offer promising solutions.This study combines Convolutional Neural Network (CNN) algorithms, Visual Geometry Group (VGG) Residual (ResNet) or VGG16 ResNet-50 specifically, image embedding alongside ML classifiers Support Vector (SVM), Artificial (ANN), Random Forest, Extreme Gradient Boosting (XGBoost).While aforementioned solutions are usually enough, real life scenarios tend differ environment conditions.This problem leads difficulty accidents process, causing users redo process even end it prematurely.To alleviate issue, this employs optimization hyperparameter tuning via Grid Search triplet loss enhance model performance.By leveraging strengths CNNs, classifiers, techniques, research aims improve accuracy efficiency processes while addressing real-world ensuring trustworthiness electronic legal documents.Evaluation conducted using ICDAR-2011 BHSig-260 datasets.Results indicate significantly improves performance SVM classification, notably elevating Area Under ROC Curve (AUC) from 0.970 0.991.

Language: Английский

---

Enhanced Land Use and Land Cover Classification Through Human Group-based Particle Swarm Optimization-Ant Colony Optimization Integration with Convolutional Neural Network

International Journal of Advanced Computer Science and Applications, Journal Year: 2023, Volume and Issue: 14(11)

Published: Jan. 1, 2023

Reliable classification of Land Use and Cover (LULC) using satellite images is essential for disaster management, environmental monitoring, urban planning. This paper introduces a unique method that combines Convolutional Neural Network (CNN) with Human Group-based Particle Swarm Optimization (HPSO) Ant Colony (ACO) algorithms to improve the accuracy LULC classification. The suggested hybrid HPSO-ACO-CNN architecture effectively solves issues feature selection, parameter optimization, model training are present in conventional techniques. During initial phases, HPSO ACO crucial identifying best hyperparameters CNN fine-tuning selection critical spectral bands. modifies CNN's (learning rate, batch size, convolutional layers), whereas finds optimal optimization technique reduces probability overfitting while substantially enhancing model's ability generalize. Utilizing selected bands optimum configuration, algorithm trained second phase. With Python implementation, this uses both spatial characteristics detects reach an outstanding 99.3% approach outperforms traditional methods like Deep (DNN), Multiclass Support Vector Machine (MSVM), Long Short-Term Memory (LSTM) experiments benchmark image datasets, demonstrating significant 10.5% increase accuracy. transforms accurate dependable classification, offering advantageous instrument remote sensing applications. It enhances area imagery evaluation by combining advantages deep learning techniques algorithms, enabling more mapping land use cover sustainable management preservation.

Language: Английский

---