Enhanced corn seed disease classification: leveraging MobileNetV2 with feature augmentation and transfer learning

Frontiers in Applied Mathematics and Statistics, Journal Year: 2024, Volume and Issue: 9

Published: Jan. 3, 2024

In the era of advancing artificial intelligence (AI), its application in agriculture has become increasingly pivotal. This study explores integration AI for discriminative classification corn diseases, addressing need efficient agricultural practices. Leveraging a comprehensive dataset, encompasses 21,662 images categorized into four classes: Broken, Discolored, Silk cut, and Pure. The proposed model, an enhanced iteration MobileNetV2, strategically incorporates additional layers—Average Pooling, Flatten, Dense, Dropout, softmax—augmenting feature extraction capabilities. Model tuning techniques, including data augmentation, adaptive learning rate, model checkpointing, dropout, transfer learning, fortify model's efficiency. Results showcase exceptional performance, achieving accuracy ~96% across classes. Precision, recall, F1-score metrics underscore proficiency, with precision values ranging from 0.949 to 0.975 recall 0.957 0.963. comparative analysis state-of-the-art (SOTA) models, outshines counterparts terms precision, F1-score, accuracy. Notably, base architecture, achieves highest values, affirming superiority accurately classifying instances within disease dataset. not only contributes growing body applications but also presents novel effective classification. robust combined competitive edge against SOTA positions it as promising solution crop management.

Language: Английский

---

Enhancing soybean classification with modified inception model: A transfer learning approach

Emirates Journal of Food and Agriculture, Journal Year: 2024, Volume and Issue: 36, P. 1 - 9

Published: April 18, 2024

The impact of deep learning (DL) is substantial across numerous domains, particularly in agriculture. Within this context, our study focuses on the classification problematic soybean seeds. dataset employed encompasses five distinct classes, totaling 5513 images. Our model, based InceptionV3 architecture, undergoes modification with addition supplementary layers to enhance efficiency and performance. Techniques such as transfer learning, adaptive rate adjustment (to 0.001), model checkpointing are integrated optimize accuracy. During initial evaluation, achieved 88.07% accuracy training 86.67% validation. Subsequent implementation tuning strategies significantly improves Augmenting architecture additional layers, including Average Pooling, Flatten, Dense, Dropout, Softmax, plays a pivotal role enhancing Evaluation metrics, precision, recall, F1-score, underscore model’s effectiveness. Precision ranges from 0.9706 1.0000, while recall values demonstrate high capture all classes. reflecting balance between precision exhibits remarkable performance ranging 0.9851 1.0000. Comparative analysis existing studies reveals competitive 98.73% by proposed model. While variations exist specific purposes datasets among studies, showcases promising seed classification, contributing advancements agricultural technology for crop health assessment management.

Language: Английский

---

Computer-aided diagnosis system for grading brain tumor using histopathology images based on color and texture features

BMC Medical Imaging, Journal Year: 2024, Volume and Issue: 24(1)

Published: July 19, 2024

Abstract Background Cancer pathology shows disease development and associated molecular features. It provides extensive phenotypic information that is cancer-predictive has potential implications for planning treatment. Based on the exceptional performance of computational approaches in field digital pathogenic, use rich images enabled us to identify low-level gliomas (LGG) from high-grade (HGG). Because differences between textures are so slight, utilizing just one feature or a small number features produces poor categorization results. Methods In this work, multiple extraction methods can extract distinct texture histopathology image data used compare classification outcomes. The successful algorithms GLCM, LBP, multi-LBGLCM, GLRLM, color moment features, RSHD have been chosen paper. LBP GLCM combined create LBGLCM. LBGLCM approach extended study scales using an pyramid, which defined by sampling both space scale. preprocessing stage first enhance contrast remove noise illumination effects. then carried out several important (texture color) images. Third, fusion reduction step put into practice decrease processed, reducing computation time suggested system. created at end categorize various brain cancer grades. We performed our analysis 821 whole-slide glioma patients Genome Atlas (TCGA) dataset. Two types included dataset: GBM LGG (grades II III). 506 315 analysis, guaranteeing representation tumor grades histopathological Results textural characteristics was validated 10-fold cross-validation technique with accuracy equals 95.8%, sensitivity 96.4%, DSC 96.7%, specificity 97.1%. combination produced significantly better accuracy, supported their synergistic significance predictive model. result indicates be objective, accurate, comprehensive prediction when paired conventional imagery. Conclusion results outperform current identifying HGG provide competitive classifying four categories literature. proposed model help stratify clinical studies, choose targeted therapy, customize specific treatment schedules.

Language: Английский

---

Differentiation of Early Sacroiliitis Using Machine-Learning- Supported Texture Analysis

Diagnostics, Journal Year: 2025, Volume and Issue: 15(2), P. 209 - 209

Published: Jan. 17, 2025

Objectives: We wished to compare the diagnostic performance of texture analysis (TA) against that a visual qualitative assessment in identifying early sacroiliitis (nr-axSpA). Methods: A total 92 participants were retrospectively included at our university hospital institution, comprising 30 controls and 62 patients with axSpA, including 32 nr-axSpA r-axSpA, who underwent MR examination sacroiliac joints. MRI 3T lumbar spine joint was performed using oblique T1-weighted (W), fluid-sensitive, fat-saturated (Fs) T2WI images. The modified New York criteria for AS used. Patients classified into group if their digital radiography (DR) and/or CT results within 7 days from showed DR grade < 2 bilateral joints or 3 unilateral joint. r-axSpA considered have confirmed diagnosis 4 thereby excluded. control healthy individuals matched terms age sex this study. First, two readers independently qualitatively scored coronal T1WI FsT2WI non-enhanced efficacies judged compared an assigned Likert score, conducting Kappa consistency test between readers. Texture models (the T1WI-TA model FsT2WI-TA model) constructed through feature extraction screening. quantitative evaluated clinical reference standard. Results: scores could significantly distinguish groups (both p 0.05). Both TA There no significant difference differential diagnoses (AUC: 0.934 vs. 0.976; = 0.1838) 0.917 0.848; 0.2592). In distinguishing groups, both superior (all (p 0.023 0.007), whereas there fsT2WI-TA 0.134 0.065). Conclusions: Based on imaging, highly effective arthritis. improved efficacy arthritis readers, while comparable

Language: Английский

---

Next-generation approach to skin disorder prediction employing hybrid deep transfer learning

Frontiers in Big Data, Journal Year: 2025, Volume and Issue: 8

Published: Feb. 19, 2025

Skin diseases significantly impact individuals' health and mental wellbeing. However, their classification remains challenging due to complex lesion characteristics, overlapping symptoms, limited annotated datasets. Traditional convolutional neural networks (CNNs) often struggle with generalization, leading suboptimal performance. To address these challenges, this study proposes a Hybrid Deep Transfer Learning Method (HDTLM) that integrates DenseNet121 EfficientNetB0 for improved skin disease prediction. The proposed hybrid model leverages DenseNet121's dense connectivity capturing intricate patterns EfficientNetB0's computational efficiency scalability. A dataset comprising 19 conditions 19,171 images was used training validation. evaluated using multiple performance metrics, including accuracy, precision, recall, F1-score. Additionally, comparative analysis conducted against state-of-the-art models such as DenseNet121, EfficientNetB0, VGG19, MobileNetV2, AlexNet. HDTLM achieved accuracy of 98.18% validation 97.57%. It consistently outperformed baseline models, achieving precision 0.95, recall 0.96, F1-score an overall 98.18%. results demonstrate the model's superior ability generalize across diverse categories. findings underscore effectiveness in enhancing classification, particularly scenarios significant domain shifts labeled data. By integrating complementary strengths provides robust scalable solution automated dermatological diagnostics.

Language: Английский

---

Oil spills and the ripple effect: exploring climate and environmental impacts through a deep learning lens

Elsevier eBooks, Journal Year: 2025, Volume and Issue: unknown, P. 279 - 289

Published: Jan. 1, 2025

Language: Английский

---

A Comparative Study of Machine Learning Techniques for Cell Annotation of scRNA-Seq Data

Algorithms, Journal Year: 2025, Volume and Issue: 18(4), P. 232 - 232

Published: April 18, 2025

Accurate cell type annotation is a critical step in single-cell RNA sequencing (scRNA-seq) analysis, enabling deeper insights into cellular heterogeneity and biological processes. In this study, we conducted comprehensive comparative evaluation of various machine learning techniques, including support vector (SVM), decision tree, random forest, logistic regression, gradient boosting, k-nearest neighbour, transformer, naive Bayes, to determine their effectiveness for annotation. These methods were evaluated using four diverse datasets comprising hundreds types across several tissues. Our results revealed that SVM consistently outperformed other emerging as the top performer three out datasets, followed closely by regression. Most demonstrated robust capabilities annotating major identifying rare populations, though Bayes was least effective due its inherent limitations handling high-dimensional interdependent data. This study provides valuable relative strengths weaknesses annotation, offering guidance selecting appropriate techniques scRNA-seq analyses.

Language: Английский

---

Advancements in Brain Tumour Analysis: A Review of Machine Learning, Deep Learning, Image Processing, and Explainable AI Techniques

Operations Research Forum, Journal Year: 2025, Volume and Issue: 6(2)

Published: May 5, 2025

Language: Английский

---

Enhancing Autism Severity Classification: Integrating LSTM into CNNs for Multisite Meltdown Grading

International Journal of Advanced Computer Science and Applications, Journal Year: 2023, Volume and Issue: 14(12)

Published: Jan. 1, 2023

Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by deficits in social interaction, verbal and non-verbal communication, often associated with cognitive neurobehavioral challenges. Timely screening diagnosis of ASD are crucial for early educational planning, treatment, family support, timely medical intervention. Manual diagnostic methods time-consuming labor-intensive, underscoring the need automated approaches to assist caretakers parents. While various researchers have employed machine learning deep techniques diagnosis, existing models fall short capturing complexity multisite meltdowns fully leveraging interdependence among these severity assessment acquired facial images children, hindering development comprehensive grading system. This paper introduces novel approach using Long Short Term Memory (LSTM) integrated Convolution Neural Network (CNN) designed identify exploit their ASD. The process begins image pre-processing, involving discrete convolution filters noise removal contrast enhancement improve quality. enhanced then undergoes instance segmentation Segment Anything model significant regions child's image. segmented region subjected principal component analysis feature extraction, features utilized LSTM-integrated CNN meltdown detection classification. trained children's extracted from videos, testing performed on videos captured during observations. Performance reveals superior results, training accuracy 88% validation 84%, outperforming conventional methods. innovative not only enhances efficiency but also provides more nuanced understanding impact severity, contributing robust

Language: Английский

---

Analisis Perbandingan Kernel Morfologi Citra Otak Ditinjau dari Nilai PSNR

Indonesian Journal of Computer Science, Journal Year: 2024, Volume and Issue: 13(3)

Published: June 30, 2024

Image processing banyak digunakan diberbagai bidang kehidupan diantaranya dibidang kedokteran untuk mendiagnosa penyakit. Salah satu penyakit yang menggunakan image adalah tumor otak. Tumor otak merupakan sangat membahayakan manusia menyerang organ Pada penelitian ini, membandingkan morfologi citra dengan kernel 11 dan 13. Data ada 5 hasil CT-Scan diproses menjadi 45 Metode metode dilasi, erosi, closing opening. Citra 13 kemudian akan dibandingkan nilai citranya MSE, RMSE PSNR. Hasil tertinggi didapat dari yaitu MSE=453.634.918, RMSE=21.298.707 PSNR=21.563739 dB. Nilai terendah dilasi MSE=9.101.720.394, RMSE=95.402.937 PSNR=8.539569 Kesimpulanya, lebih bagus hasilnya jika ditinjau

---