A comparison review of transfer learning and self-supervised learning: Definitions, applications, advantages and limitations

Expert Systems with Applications, Journal Year: 2023, Volume and Issue: 242, P. 122807 - 122807

Published: Dec. 2, 2023

Deep learning has emerged as a powerful tool in various domains, revolutionising machine research. However, one persistent challenge is the scarcity of labelled training data, which hampers performance and generalisation deep models. To address this limitation, researchers have developed innovative methods to overcome data enhance model capabilities. Two prevalent techniques that gained significant attention are transfer self-supervised learning. Transfer leverages knowledge learned from pre-training on large-scale dataset, such ImageNet, applies it target task with limited data. This approach allows models benefit representations effectively new tasks, resulting improved generalisation. On other hand, focuses using pretext tasks do not require manual annotation, allowing them learn valuable large amounts unlabelled These can then be fine-tuned for downstream mitigating need extensive In recent years, found applications fields, including medical image processing, video recognition, natural language processing. approaches demonstrated remarkable achievements, enabling breakthroughs areas disease diagnosis, object understanding. while these offer numerous advantages, they also limitations. For example, may face domain mismatch issues between requires careful design ensure meaningful representations. review paper explores fields within past three years. It delves into advantages limitations each approach, assesses employing techniques, identifies potential directions future By providing comprehensive current methods, article offers guidance selecting best technique specific issue.

Language: Английский

---

Plant disease detection and classification techniques: a comparative study of the performances

Journal Of Big Data, Journal Year: 2024, Volume and Issue: 11(1)

Published: Jan. 2, 2024

Abstract One of the essential components human civilization is agriculture. It helps economy in addition to supplying food. Plant leaves or crops are vulnerable different diseases during agricultural cultivation. The halt growth their respective species. Early and precise detection classification may reduce chance additional damage plants. these have become serious problems. Farmers’ typical way predicting classifying plant leaf can be boring erroneous. Problems arise when attempting predict types manually. inability detect classify quickly result destruction crop plants, resulting a significant decrease products. Farmers that use computerized image processing methods fields losses increase productivity. Numerous techniques been adopted applied based on images infected crops. Researchers made progress past by exploring various techniques. However, improvements required as reviews, new advancements, discussions. technology significantly production all around world. Previous research has determined robustness deep learning (DL) machine (ML) such k-means clustering (KMC), naive Bayes (NB), feed-forward neural network (FFNN), support vector (SVM), k-nearest neighbor (KNN) classifier, fuzzy logic (FL), genetic algorithm (GA), artificial (ANN), convolutional (CNN), so on. Here, from DL ML included this particular study, CNNs often favored choice for due inherent capacity autonomously acquire pertinent features grasp spatial hierarchies. Nevertheless, selection between conventional hinges upon problem, accessibility data, computational capabilities accessible. Accordingly, numerous advanced tasks, DL, mainly through CNNs, preferred ample data resources available show good effects datasets, but not other datasets. Finally, paper, author aims keep future researchers up-to-date with performances, evaluation metrics, results previously used forms using image-processing intelligence (AI) field.

Language: Английский

---

CerCan·Net: Cervical cancer classification model via multi-layer feature ensembles of lightweight CNNs and transfer learning

Expert Systems with Applications, Journal Year: 2023, Volume and Issue: 229, P. 120624 - 120624

Published: June 2, 2023

Language: Английский

---

Skin-CAD: Explainable deep learning classification of skin cancer from dermoscopic images by feature selection of dual high-level CNNs features and transfer learning

Computers in Biology and Medicine, Journal Year: 2024, Volume and Issue: 178, P. 108798 - 108798

Published: June 25, 2024

Language: Английский

---

Detection and identification of plant leaf diseases using YOLOv4

Frontiers in Plant Science, Journal Year: 2024, Volume and Issue: 15

Published: April 22, 2024

Detecting plant leaf diseases accurately and promptly is essential for reducing economic consequences maximizing crop yield. However, farmers’ dependence on conventional manual techniques presents a difficulty in pinpointing particular diseases. This research investigates the utilization of YOLOv4 algorithm detecting identifying study uses comprehensive Plant Village Dataset, which includes over fifty thousand photos healthy diseased leaves from fourteen different species, to develop advanced disease prediction systems agriculture. Data augmentation including histogram equalization horizontal flip were used improve dataset strengthen model’s resilience. A assessment was conducted, involved comparing its performance with established target identification methods Densenet, Alexanet, neural networks. When dataset, it achieved an impressive accuracy 99.99%. The evaluation criteria, accuracy, precision, recall, f1-score, consistently showed high value 0.99, confirming effectiveness proposed methodology. study’s results demonstrate substantial advancements detection underscore capabilities as sophisticated tool accurate prediction. These developments have significant significance everyone agriculture, researchers, farmers, providing improved capacities control protection.

Language: Английский

---

Cervical Cancer Diagnosis Based on Multi-Domain Features Using Deep Learning Enhanced by Handcrafted Descriptors

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(3), P. 1916 - 1916

Published: Feb. 2, 2023

Cervical cancer, among the most frequent adverse cancers in women, could be avoided through routine checks. The Pap smear check is a widespread screening methodology for timely identification of cervical but it susceptible to human mistakes. Artificial Intelligence-reliant computer-aided diagnostic (CAD) methods have been extensively explored identify cancer order enhance conventional testing procedure. In attain remarkable classification results, current CAD systems require pre-segmentation steps extraction cells from pap slide, which complicated task. Furthermore, some models use only hand-crafted feature cannot guarantee sufficiency phases. addition, if there are few data samples, such as cell datasets, deep learning (DL) alone not perfect choice. existing obtain attributes one domain, integration features multiple domains usually increases performance. Hence, this article presents model based on extracting domain. It does process thus less complex than methods. employs three compact DL high-level spatial rather utilizing an individual with large number parameters and layers used CADs. Moreover, retrieves several statistical textural descriptors including time–frequency instead employing single domain demonstrate clearer representation features, case examines influence each set handcrafted accuracy independently hybrid. then consequences combining obtained CNN combined features. Finally, uses principal component analysis merge entire investigate effect merging numerous various results. With 35 components, achieved by quatric SVM proposed reached 100%. performance described proves that able boost accuracy. Additionally, comparative analysis, along other present studies, shows competing capacity CAD.

Language: Английский

---

A Federated Learning CNN Approach for Tomato Leaf Disease with Severity Analysis

Published: Aug. 23, 2023

This research includes four disease levels and one healthy level in a federated learning Convolutional Neural Network (CNN) model for detecting categorizing tomato leaf illnesses across five severity levels. Data from customers were used to analyze the model, with each client's performance measures precision, recall, F1-score, accuracy being reviewed. The obtained results show that consistently generated of good quality, an range 96% 98% all Client 4 had greatest Class 1 (healthy), but Clients 2 3 achieved same every class, suggesting datasets or data distribution comparable. 5 (disease 4) highest recall clients, demonstrating model's skill identifying this illness. An 97% was shown when assessed using locally averaged global values parameters clients. Additionally, study has contrasted clients macro average, weighted micro average averaging techniques. These highlighted potential usefulness agricultural settings by consistent utilizing techniques customers. While protecting privacy, CNN accurately identifies diseases at various levels, allowing targeted practices interventions reduce yield loss improve fruit quality.

Language: Английский

---

EfficientRMT-Net—An Efficient ResNet-50 and Vision Transformers Approach for Classifying Potato Plant Leaf Diseases

Sensors, Journal Year: 2023, Volume and Issue: 23(23), P. 9516 - 9516

Published: Nov. 30, 2023

The primary objective of this study is to develop an advanced, automated system for the early detection and classification leaf diseases in potato plants, which are among most cultivated vegetable crops worldwide. These diseases, notably late blight caused by Alternaria solani Phytophthora infestans, significantly impact quantity quality global production. We hypothesize that integration Vision Transformer (ViT) ResNet-50 architectures a new model, named EfficientRMT-Net, can effectively accurately identify various diseases. This approach aims overcome limitations traditional methods, often labor-intensive, time-consuming, prone inaccuracies due unpredictability disease presentation. EfficientRMT-Net leverages CNN model distinct feature extraction employs depth-wise convolution (DWC) reduce computational demands. A stage block structure also incorporated improve scalability sensitive area detection, enhancing transferability across different datasets. tasks performed using average pooling layer fully connected layer. was trained, validated, tested on custom datasets specifically curated detection. EfficientRMT-Net's performance compared with other deep learning transfer techniques establish its efficacy. Preliminary results show achieves accuracy 97.65% general image dataset 99.12% specialized Potato dataset, outperforming existing methods. demonstrates high level proficiency correctly classifying identifying even cases distorted samples. provides efficient accurate solution plant potentially enabling farmers enhance crop yield while optimizing resource utilization. confirms our hypothesis, showcasing effectiveness combining ViT addressing complex agricultural challenges.

Language: Английский

---

TL-YOLOv8: A Blueberry Fruit Detection Algorithm Based on Improved YOLOv8 and Transfer Learning

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 86378 - 86390

Published: Jan. 1, 2024

Language: Английский

---

Transfer learning in agriculture: a review

Artificial Intelligence Review, Journal Year: 2025, Volume and Issue: 58(4)

Published: Jan. 25, 2025

Language: Английский

---