Deep Learning Models Research Articles

Detection and Classification of Agave angustifolia Haw Using Deep Learning Models

In Oaxaca, Mexico, there are more than 30 species of the Agave genus, and its cultivation is of great economic and social importance. The incidence of pests, diseases, and environmental stress cause significant losses to the crop. The identification of damage through non-invasive tools based on visual information is important for reducing economic losses. The objective of this study was to evaluate and compare five deep learning models: YOLO versions 7, 7-tiny, and 8, and two from the Detectron2 library, Faster-RCNN and RetinaNet, for the detection and classification of Agave angustifolia plants in digital images. In the town of Santiago Matatlán, Oaxaca, 333 images were taken in an open-air plantation, and 1317 plants were labeled into five classes: sick, yellow, healthy, small, and spotted. Models were trained with a 70% random partition, validated with 10%, and tested with the remaining 20%. The results obtained from the models indicate that YOLOv7 is the best-performing model, in terms of the test set, with a mAP of 0.616, outperforming YOLOv7-tiny and YOLOv8, both with a mAP of 0.606 on the same set; demonstrating that artificial intelligence for the detection and classification of Agave angustifolia plants under planting conditions is feasible using digital images.

Using deep learning and word embeddings for predicting human agreeableness behavior

The latest advancements of deep learning have resulted in a new era of natural language processing. The machines now possess an unparallel ability to interpret and engage with various tasks such as text classification, content generation and natural language understanding. This development extended to the analysis of human behavior, where deep learning models are used to decode human personality. Due to the rise of social media, generating huge amounts of textual data that reshaped communication patterns. Understanding personality traits is a challenging topic which helps us to explore the patterns of thoughts, feelings and behaviors which are helpful for recruitment, career counselling and consumers’ behavior for marketing, etc. In this research study, the main aim is to predict the human personality trait of agreeableness showing whether a person is emotional who feels a lot or thinker who is logical and has rational thinking. This behavior leads to analyzing them as cooperative, friendly and respecting difference of views. For comprehensive empirical analysis, shallow machine learning models, ensemble models, and deep learning technique including state of the art transformer-based models are applied on real-world dataset of MBTI. For feature engineering, textual features of TF-IDF and POS tagging and word embeddings such as word2vec, glove and sentence embeddings are explored. The results analysis shows the highest performance 91.57% with sentence embeddings utilizing Bi-LSTM algorithm that highlights the power of this study as compared to existing studies in the relevant literature.

Enhanced Diagnosis of Lung Cancer through an Ensemble Learning Model leveraging an Adaptive Optimization Algorithm

Early and accurate diagnosis of lung cancer is crucial to improving patient outcomes and survival rates. Machine and deep learning models have emerged as promising tools to improve the accuracy and efficiency of disease diagnosis. However, achieving optimal diagnostic performance remains a challenging task in medical research. This study integrates ensemble learning techniques with an adaptive optimization algorithm to enhance the accuracy of lung cancer diagnosis. By combining the predictive potential of multiple base classifiers, the ensemble-learning model improves overall performance and mitigates the weaknesses of individual classifiers. Additionally, the adaptive optimization algorithm dynamically adjusts the model parameters to optimize the classification performance. The effectiveness of the approach was evaluated using a comprehensive dataset that includes lung cancer images. Rigorous evaluation and comparison with state-of-the-art models showed that the proposed method achieved superior diagnostic performance, reaching an overall accuracy of 99%.

Dynamic Arithmetic Optimization Algorithm with Deep Learning-based Intrusion Detection System in Wireless Sensor Networks

A Wireless Sensor Network (WSN) encompasses interconnected Sensor Nodes (SNs) that interact wirelessly to collect and transfer data. Security in the context of WNS refers to protocols and measures implemented for the overall functionality of the network, along with protecting the availability, confidentiality, and integrity of data against tampering, unauthorized access, and other possible security risks. An Intrusion Detection System (IDS) utilizing Deep Learning (DL) and Feature Selection (FS) leverages advanced methods to enhance effectiveness in the detection of malicious activities in a network by enhancing relevant data features and leveraging the power of Deep Neural Networks (DNNs). This study presents a Dynamic Arithmetic Optimization Algorithm within a DL-based IDS (DAOADL-IDS) in WSNs. The purpose of DAOADL-IDS is to recognize and classify intrusions in a WSN using a metaheuristic algorithm and DL models. To accomplish this, the DAOADL-IDS technique utilizes a Z-score data normalization approach to resize the input dataset in a compatible format. In addition, DAOADL-IDS employs a DAOA-based FS (DAOA-FS) model to select an optimum set of features. A Stacked Deep Belief Network (SDBN) model is employed for the Intrusion Detection (ID) process. The hyperparameter selection of the SDBN model is accomplished using the Bird Swarm Algorithm (BSA). A wide experimental analysis of the proposed DAOADL-IDS method was performed on a benchmark dataset. The performance validation of the DAOADL-IDS technique showed an accuracy of 99.68%, demonstrating superior performance over existing techniques under various measures.

Evaluating Deep Learning Models for Website Phishing Attack Detection: A Comparative Analysis

Phishing attacks remain a significant security threat in cyberspace, targeting individuals and businesses to steal confidential information. Traditional detection methods often struggle to identify newly created or altered phishing sites, highlighting the need for more adaptive solutions. This study evaluates the performance of various deep learning (DL) models for detecting online phishing attacks. A comparative analysis of single and hybrid DL models, including CNN, LSTM, BiGRU, and their combinations, is conducted. The evaluation is based on metrics such as accuracy, precision, recall, and F1-score, derived from 17 peer-reviewed publications published between 2019 and 2024. Results indicate that hybrid models, particularly ODAE-WPDC, exhibit superior performance, achieving accuracy rates of up to 99.28% and robust results across all metrics. Single models, such as CNN and BiGRU, also demonstrate strong performance, with accuracy ranging from 97% to 99.5%. This research underscores the efficacy of deep learning architectures in phishing detection and offers practical guidance for selecting optimal models based on specific requirements.

Polymer Concretes Based on Various Resins: Modern Research and Modeling of Mechanical Properties

This review is devoted to experimental studies and modeling in the field of mechanical and physical properties of polymer concretes and polymer-modified concretes. The review analyzes studies carried out over the past two years. The paper examines the properties of polymer concretes based on various polymer resins and presents the advantages and disadvantages of various models developed to predict the mechanical properties of materials. Based on data in the literature, the most promising polymers for use in the field of road surface repair are polymer concretes with poly(meth)acrylic resins. It was found that the most adequate and productive models are the deep machine learning model—using several hidden layers that perform calculations based on input parameters—and the extreme gradient boosting model. In particular, the extreme gradient boosting model showed high R2 values in forecasting (in the range of 0.916–0.981) when predicting damping coefficient and ultimate compressive strength. In turn, among the additives to Portland cement concrete, the most promising are natural polymers, such as mammalian gelatin and cold fish gelatin, and superabsorbent polymers. These additives allow for an improvement in compressive strength of 200% or more. The review may be of interest to engineers specializing in building construction, materials scientists involved in the development and implementation of new materials into production, as well as researchers in the interdisciplinary fields of chemistry and technology.

Retinal Image Augmentation using Composed GANs

Medical image analysis faces a significant challenge in the scarcity of annotated data, which is crucial for developing generalizable Deep Learning (DL) models that require extensive training data. Consequently, the field of medical image generation has garnered substantial interest and potential for further exploration. Besides widely employed data augmentation techniques, such as rotation, reflection, and scaling, Generative Adversarial Networks (GANs) have demonstrated the ability to effectively leverage additional information from datasets by generating synthetic samples from real images. In the context of retinal image synthesis, an image-to-image translation approach is frequently adopted to generate retinal images from available vessel maps, which can be scarce and resource-intensive to obtain. Deviating from prior work reliant on pre-existing vessel maps, this study proposes a learning-based model that is independent of vessel maps, utilizing Progressive Growing GAN (PGGAN) to generate vascular networks from random noise. The visual and quantitative evaluations conducted suggest that the majority of the images generated by the proposed model are substantially distinct from the training set while maintaining a high proportion of true image quality, underscoring the model's potential as a powerful tool for data augmentation.

Probabilistic regression for autonomous terrain relative navigation via multi-modal feature learning

The extension of human spaceflight across an ever-expanding domain, in conjunction with intricate mission architectures demands a paradigm shift in autonomous navigation algorithms, especially for the powered descent phase of planetary landing. Deep learning architectures have previously been explored to perform low-dimensional localization with limited success. Due to the expectations regarding novel algorithms in the context of real missions, the proposed approaches must be rigorously evaluated in extraneous scenarios and demonstrate sufficient robustness. In the current work, a novel formulation is proposed to train CNN-based Deep Learning (DL) models in a multi-layer cascading architecture and utilize the resulting classification probabilities as regression weights to estimate the position of the lander spacecraft. The approach leverages image intensity and depth data provided by multiple sensors on board to accurately determine the spacecraft’s location relative to the observed terrain at a specific altitude. Navigation performance is validated through Monte Carlo analysis, demonstrating the efficacy of the proposed DL architecture and the subsequent state-estimation framework across several simulated scenarios. It shows tremendous promise in extending the multi-modal feature learning approach to realistic missions.

Deep Multiple Instance Learning Model to Predict Outcome of Pancreatic Cancer Following Surgery

Background/Objectives: Pancreatic ductal adenocarcinoma (PDAC) is a cancer with very poor prognosis despite early surgical management. To date, only clinical variables are used to predict outcome for decision-making about adjuvant therapy. We sought to generate a deep learning approach based on hematoxylin and eosin (H&E) or hematoxylin, eosin and saffron (HES) whole slides to predict patients’ outcome, compare these new entities with known molecular subtypes and question their biological significance; Methods: We used as a training set a retrospective private cohort of 206 patients treated by surgery for PDAC cancer and a validation cohort of 166 non-metastatic patients from The Cancer Genome Atlas (TCGA) PDAC project. We estimated a multi-instance learning survival model to predict relapse in the training set and evaluated its performance in the validation set. RNAseq and exome data from the TCGA PDAC database were used to describe the transcriptomic and genomic features associated with deep learning classification; Results: Based on the estimation of an attention-based multi-instance learning survival model, we identified two groups of patients with a distinct prognosis. There was a significant difference in progression-free survival (PFS) between these two groups in the training set (hazard ratio HR = 0.72 [0.54;0.96]; p = 0.03) and in the validation set (HR = 0.63 [0.42;0.94]; p = 0.01). Transcriptomic and genomic features revealed that the poor prognosis group was associated with a squamous phenotype. Conclusions: Our study demonstrates that deep learning could be used to predict PDAC prognosis and offer assistance in better choosing adjuvant treatment.

Sentiment Analysis for E-commerce Product Reviews by Deep Learning Model of Bert- BiGRU-Softmax

The sentiment analysis has the key problem for the e-commerce product quality management, which customers know their people’s attitudes on interested products by e-commerce reviews. Meanwhile, manufacturers are able to learn the public sentiment on their products being sold in E-commerce platforms. This paper proposed the deep learning models of Bert-BiGRU-Softmax, which uses the input layer of Sentiment Bert model to extract multi-dimensional of E-commerce reviews, the hidden layer of Bidirectional GRU model to obtain semantic codes and calculate representing weights of reviews, the Softmax with attention mechanism as the output layer to classify the sentiment tendency. We conduct experiments on a large-scale dataset involving over 500 thousand reviews compared with different learning models. The experimental results show that the proposed models reach the high accuracy 95.5% on the E-commerce reviews, and the outperforms of RNN, BiGRU, Bert-BiLSTM in terms of accuracy and loss.

DeepYoga: Enhancing Practice with a Real-Time Yoga Pose Recognition System

The adoption of yoga as a holistic wellness practice is increasing throughout the world. However, in the absence of a personalized expert, especially in an online environment, there is a need for reliable and accurate methods for yoga posture recognition. Maintaining correct yoga postures is essential to reap holistic health benefits in the long term and address chronic medical issues. This paper presents DeepYoga, a novel approach to improve posture recognition accuracy with the support of deep learning models. The proposed approach uses a dataset of accurate yoga pose images encompassing five distinct poses. Landmarks extracted from the practitioner's body in the images are then used to train a Convolutional Neural Network (CNN) for accurate pose classification. The trained model is then used to detect yoga poses from real-time videos of yoga practitioners. Then, the system provides users with real-time feedback and visual suggestions, helping them improve physical alignment and reduce the risk of injury. The proposed method achieved an overall high accuracy of 99.02% in pose detection while trying to minimize the use of resources as much as possible to make it more accessible.

Deep learning model for efficient traffic forecasting in intelligent transportation systems

Deep learning model for efficient traffic forecasting in intelligent transportation systems

Deep Learning-Based Control System Design for Emergency Vehicles through Intersections

This paper addresses the challenge of integrating priority passage for emergency vehicles with optimal intersection control in modern urban traffic. It proposes an innovative strategy based on deep learning to enable emergency vehicles to pass through intersections efficiently and safely. The research aims to develop a deep learning model that utilizes intersection violation monitoring cameras to identify emergency vehicles in real time. This system adjusts traffic signals to ensure the rapid passage of emergency vehicles while simultaneously optimizing the overall efficiency of the traffic system. In this study, OpenCV is used in combination with Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) to jointly complete complex image processing and analysis tasks, to realize the purpose of fast travel of emergency vehicles. At the end of this study, the principle of the You Only Look Once (YOLO) algorithm can be used to design a website and a mobile phone application (app) to enable private vehicles with emergency needs to realize emergency passage through the application, which is also of great significance to improve the overall level of urban traffic management, reduce traffic congestion and promote the development of related technologies.

An Earth Mover's Distance-Based Self-Supervised Framework for Cellular Dynamic Grading in Live-Cell Imaging.

Cellular appearance and its dynamics frequently serve as a proxy measurement of live-cell physiological properties. The computational analysis of cell properties is considered to be a significant endeavor in biological and biomedical research. Deep learning has garnered considerable success across various fields. In light of this, various neural networks have been developed to analyze live-cell microscopic videos and capture cellular dynamics with biological significance. Specifically, cellular dynamic grading (CDG) is the task that provides a predefined dynamic grade for a live-cell according to the speed of cellular deformation and intracellular movement. This task involves recording the morphological and cytoplasmic dynamics in live-cell microscopic videos. Similar to other medical image processing tasks, CDG faces challenges in collecting and annotating cellular videos. These deficiencies in medical data limit the performance of deep learning models. In this article, we propose a novel self-supervised framework to overcome these limitations for the CDG task. Our framework relies on the assumption that increasing or decreasing cell dynamic grades is consistent with accelerating or decelerating cell appearance change in videos, respectively. This consistency is subsequently incorporated as a constraint in the loss function for the self-supervised training strategy. Our framework is implemented by formulating a probability transition matrix based on the Earth Mover's Distance and imposing a loss constraint on the elements of this matrix. Experimental results demonstrate that our proposed framework enhances the model's ability to learn spatiotemporal dynamics. Furthermore, our framework outperforms the existing methods on our cell video database.

Optimizing Stroke Classification with Pre-Trained Deep Learning Models

Background/Objectives: Insufficient blood supply to the brain, whether due to blocked arteries (ischemic stroke) or bleeding (hemorrhagic stroke), leads to brain cell death and cognitive impairment. Ischemic strokes, which are more common, occur when blood flow to the brain is obstructed. Magnetic resonance imaging (MRI) scans are essential for distinguishing stroke types, but precise and timely identification of ischemic strokes is crucial for effective treatment. Manual diagnosis can be difficult due to high patient volumes and time constraints in hospitals. This study aims to investigate the use of deep learning techniques for predicting ischemic strokes with high accuracy, enabling earlier diagnosis and intervention. Methods: The study utilized advanced deep learning algorithms, specifically ConvNeXt Base, to analyze large datasets of medical imaging data, focusing on MRI scans. The model was trained and validated on a labeled dataset to identify critical indicators and patterns associated with stroke risk. The performance of the model was evaluated based on accuracy metrics to determine its predictive capabilities. Results: ConvNeXt Base achieved an overall accuracy of 84% on the validation set, demonstrating its effectiveness in identifying ischemic strokes. The model was able to detect key patterns linked to stroke risk, highlighting its potential for use in clinical settings to aid in early diagnosis and decision-making. Conclusions: ConvNeXt Base reveals promise in improving stroke prediction accuracy, enabling earlier diagnosis and personalized treatment, which could lead to faster, more effective medical interventions.

Unveiling wind-thrown trees: Detection and quantification of wind-thrown tree stems on UAV-orthomosaics based on UNet and a heuristic stem reconstruction

Severe storm events cause significant biomass losses in European forests. In addition to direct storm damage, biotic, abiotic, and market-related factors contribute to further damage with far-reaching effects on forest and nature conservation. These subsequent damages can be minimized if the amount of damaged timber and its spatial distribution are known and can be used to plan salvage operations. Traditional methods and satellite-based remote sensing can only delineate affected areas and estimate the amount of damaged timber. Aerial remote sensing can determine the spatial distribution of tree trunks with a detection rate of up to 92%, but it cannot quantify the volume of individual trunks. Newer UAV approaches are promising, but a methodology for quantifying fallen trees has not yet been published. This work addresses this gap by using UAVs orthomosaics and a hybrid heuristic and deep learning approach to capture the spatial distribution of tree trunks and estimate their volume while even reconstructing occluded stem parts. For the training of the deep learning model, high-resolution orthomosaics were used, in which 760 trunks were manually outlined and labeled. The method achieved an average stem detection rate DR25 of 92.6%. while an average classification error rate (CE) of 1.6% and a reconstruction error rate of 0.2% were reported. In the consecutive quantification of the stem volume, a relative bias rbias of −4.35% was reported for 560 reference stems at 15 investigation sites, whereas the result differ between the site (rbias= [−22.44%, 9.06%]) due weather conditions during the flight, stand parameters and the wind-throw pattern itself. The proposed method provides additional information for the management of wind-throw areas and the monitoring of biomass and carbon cycles, supports the understanding of wind-throw dynamics, and promotes the development of sustainable management strategies.

An enhanced lightweight T-Net architecture based on convolutional neural network (CNN) for tomato plant leaf disease classification.

Tomatoes are a widely cultivated crop globally, and according to the Food and Agriculture Organization (FAO) statistics, tomatoes are the third after potatoes and sweet potatoes. Tomatoes are commonly used in kitchens worldwide. Despite their popularity, tomato crops face challenges from several diseases, which reduce their quality and quantity. Therefore, there is a significant problem with global agricultural productivity due to the development of diseases related to tomatoes. Fusarium wilt and bacterial blight are substantial challenges for tomato farming, affecting global economies and food security. Technological breakthroughs are necessary because existing disease detection methods are time-consuming and labor-intensive. We have proposed the T-Net model to find a rapid, accurate approach to tackle the challenge of automated detection of tomato disease. This novel deep learning model utilizes a unique combination of the layered architecture of convolutional neural networks (CNNs) and a transfer learning model based on VGG-16, Inception V3, and AlexNet to classify tomato leaf disease. Our suggested T-Net model outperforms earlier methods with an astounding 98.97% accuracy rate. We prove the effectiveness of our technique by extensive experimentation and comparison with current approaches. This study offers a dependable and understandable method for diagnosing tomato illnesses, marking a substantial development in agricultural technology. The proposed T-Net-based framework helps protect crops by providing farmers with practical knowledge for managing disease. The source code can be accessed from the given link.

Eye Movement Classification using Feature Engineering and Ensemble Machine Learning

This paper explores the classification of gaze direction using electrooculography (EOG) signals, integrating signal processing, deep learning, and ensemble learning techniques to enhance accuracy and reliability. A complex technique is proposed in which several feature types are derived from EOG data. Spectral properties generated from power spectral density analysis augment basic statistical characteristics such as mean and standard deviation, revealing the frequency content of the signal. Skewness, kurtosis, and cross-channel correlations are also used to represent intricate nonlinear dynamics and inter-channel interactions. These characteristics are then reformatted into a two-dimensional array imitating picture data, enabling the use of the pre-trained ResNet50 model to extract deep and high-level characteristics. Using these deep features, an ensemble of bagging-trained decision trees classifies gaze directions, lowering model variance and increasing prediction accuracy. The results show that the ensemble deep learning model obtained outstanding performance metrics, with accuracy and sensitivity ratings exceeding 97% and F1-score of 98%. These results not only confirm the effectiveness of the proposed approach in managing challenging EOG signal classification tasks but also imply important consequences for the improvement of Human-Computer Interaction (HCI) systems, especially in assistive technologies where accurate gaze tracking is fundamental.

Deep learning-assisted colonoscopy images for prediction of mismatch repair deficiency in colorectal cancer.

Deficient mismatch repair or microsatellite instability is a major predictive biomarker for the efficacy of immune checkpoint inhibitors of colorectal cancer. However, routine testing has not been uniformly implemented due to cost and resource constraints. We developed and validated a deep learning-based classifiers to detect mismatch repair-deficient status from routine colonoscopy images. We obtained the colonoscopy images from the imaging database at Endoscopic Center of the Sixth Affiliated Hospital, Sun Yat-sen University. Colonoscopy images from a prospective trial (Neoadjuvant PD-1 blockade by toripalimab with or without celecoxib in mismatch repair-deficient or microsatellite instability-high locally advanced colorectal cancer) were used to test the model. A total of 5226 eligible images from 892 tumors from the consecutive patients were utilized to develop and validate the deep learning model. 2105 colorectal cancer images from 306 tumors were randomly selected to form model development dataset with a class-balanced approach. 3121 images of 488 proficient mismatch repair tumors and 98 deficient mismatch repair tumors were used to form the independent dataset. The model achieved an AUROC of 0.948 (95% CI 0.919-0.977) on the test dataset. On the independent validation dataset, the AUROC was 0.807 (0.760-0.854), and the NPV in was 94.2% (95% CI 0.918-0.967). On the prospective trial dataset, the model identified 29 tumors among the 33 deficient mismatch repair tumors (87.88%). The model achieved a high NPV in detecting deficient mismatch repair colorectal cancers. This model might serve as an automatic screening tool.

A Hybrid Metaheuristic Aware Enhanced Deep Learning Approach for Software Effort Estimation

Software Effort Estimating (SEE) is a fundamental task in all software development lifecycles and procedures. Therefore, when deciding how to anticipate effort in a variety of project types, the comparative assessment of effort prediction methods has emerged as a standard strategy. Unfortunately, these studies include a range of sample techniques and error metrics, making a comparison with other work challenging. To overcome these drawbacks, this study proposes a deep learning model to effectively estimate software effort. The estimation is mainly focused on minimizing the cost and time consumption. The input data is taken from the dataset and preprocessing is performed to remove the noise content. Then the required features are extracted using the preprocessed data with the help of the simple and higher-order statistical features. A novel Modified Chaotic Enriched Jaya with Moth Flame Optimization (MCEJMO) algorithm is introduced for feature selection to enhance SEE accuracy. The estimation is performed using Multilayer Long Short-Term Memory (M-LSTM). The proposed method achieved a Mean Square Error (MSE) of 0.2825 for dataset 1 and 0.2285 for dataset 2.