ImageNet Weights Research Articles

Multiclass Classification of Tomato Leaf Diseases Using Convolutional Neural Networks and Transfer Learning

ABSTRACTTomato (biological name: Solanum lycopersicum) is an important food crop worldwide. However, due to climatic changes and various diseases, the yield of tomatoes decreased significantly, being detrimental from an economic point of view. Various diseases infect the tomato leaves, such as bacterial and septorial leaf spots, early blight and mosaic virus, to name a few. If uncared, these tomato leaf diseases (TLDs) can spread to other leaves and the fruit. Hence it is vital to detect these diseases as early as possible. Leaf examination is one of the standard techniques to identify and control the spread of diseases. Big Data has made substantial progress, and with the help of computer vision and deep learning techniques to analyse data, we can identify the diseased leaves and help control the disease's spread further. This study used three lightweight midgeneration convolutional neural networks (CNNs) classification network architectures which has the scope to be deployed in IoT devices to help the agricultural community tackle TLDs. It also shows the efficacy of the models with and without geometric data augmentation. The model was trained on a Kaggle data set containing a more significant number of samples to make a robust model aware of broader data distribution and validated on the Plant Village dataset to test its efficacy. The results show that applying transfer learning using ImageNet weights to the MobileNet Architecture using geometrically augmented sample images yields a train and test accuracy of 99.71% and 99.49%, respectively.

Enhancing COVID-19 disease severity classification through advanced transfer learning techniques and optimal weight initialization schemes

ObjectiveMedical imaging plays a central role in medicine today by supporting diagnosis and treatment. For small medical image datasets, training from scratch is a time-consuming process and transfer learning emerges as a solution. In these cases, ImageNet weights are usually used as initial weights, after which fine-tuning is performed. We propose a methodology for COVID-19 severity classification (mild, moderate, severe, critical) in chest X-ray images based on transfer learning using DenseNet121 architecture as a base model. MethodsThe novelty of the approach mainly lies in the investigation of three different weight initialization schemes (i) ImageNet (ii) CheXNeXt (iii) DeepCOVID-XR, which are similar in a graded manner in terms of image nature, with ImageNet being the least similar, CheXNeXt similar as they were trained on different lung diseases not including COVID-19 and DeepCOVID-XR the most similar as they were obtained by training specifically on COVID-19 X-ray images. ResultsThe results show that the worst results are achieved using ImageNet as initial weights (average AUC = 0.700), followed by the better results with DeepCOVID-XR (average AUC = 0.774) while CheXNeXt performed significantly better (average AUC = 0.917). The results of weaker performing classification models were improved when the severe and critical classes were merged, to account for the similarity between these classes in the dataset (average AUC for initialization schemes were ImageNet AUC = 0.867, CheXNeXt AUC = 0.900 and DeepCOVID-XR AUC = 0.794). ConclusionThis leads to a conclusion that, in the medical domain, where image datasets are usually small and highly imbalanced, if initial weights are chosen to be in nature similar to the new dataset, achieved results are better. However, there is no need to start from weights obtained during training on the same disease, as this may cause overfitting. There are significant discrepancies between ImageNet and medical imaging datasets and the results from this paper could help in guiding future implementations of transfer learning in medical applications.

Advances in Deep Learning for Skin Cancer Diagnosis

The most prevalent type of cancer worldwide is known as skin cancer. Early detection is critical because if left undiagnosed in the primary stage, it might be fatal. Although there are differences within the class and high inter-class similarities, it is too difficult to distinguish with the naked eye. Owing to the disease's global prevalence, a number of deep learning based automated systems were created thus far to help doctors identify skin lesions early on. Using pre-trained ImageNet weights and fine-tuning the Convolutional Neural Networks (CNNs), we trained VGG19 on the HAM10000 dataset. The optimal performance was observed with FT. The model that was created, which yielded an accuracy that was greater overall than the one used in transfer learning, was 82.4±1.9 %. By offering a second opinion and supporting the clinician's diagnosis, this performance could lower morbidity and treatment costs.

DentAge: Deep learning for automated age prediction using panoramic dental X-ray images.

Age estimation plays a crucial role in various fields, including forensic science and anthropology. This study aims to develop and validate DentAge, a deep-learning model for automated age prediction using panoramic dental X-ray images. DentAge was trained on a dataset comprising 21,007 panoramic dental X-ray images sourced from a private dental center in Slovenia. The dataset included subjects aged 4 to 97 years with various dental conditions. Transfer learning was employed, initializing the model with ImageNet weights and fine-tuning on the dental image dataset. The model was trained using stochastic gradient descent with momentum, and mean absolute error (MAE) served as the objective function. Across the test dataset, DentAge achieved an MAE of 3.12 years, demonstrating its efficacy in age prediction. Notably, the model performed well across different age groups, with MAEs ranging from 1.94 (age group [10-20]) to 13.40 years (age group [90-100]). Visual evaluation revealed factors contributing to prediction errors, including prosthetic restorations, tooth loss, and bone resorption. DentAge represents a significant advancement in automated age prediction within dentistry. The model's robust performance across diverse age groups and dental conditions underscores its potential utility in real-world scenarios. Our model will be accessible to the public for further adjustments and validation, ensuring DentAge's effectiveness and trustworthiness in practical scenarios.

Self-supervised in-domain representation learning for remote sensing image scene classification

Transferring the ImageNet pre-trained weights to the various remote sensing tasks has produced acceptable results and reduced the need for labeled samples. However, the domain differences between ground imageries and remote sensing images cause the performance of such transfer learning to be limited. The difficulty of annotating remote sensing images is well-known as it requires domain experts and more time, whereas unlabeled data is readily available. Recently, self-supervised learning, which is a subset of unsupervised learning, emerged and significantly improved representation learning. Recent research has demonstrated that self-supervised learning methods capture visual features that are more discriminative and transferable than the supervised ImageNet weights. We are motivated by these facts to pre-train the in-domain representations of remote sensing imagery using contrastive self-supervised learning and transfer the learned features to other related remote sensing datasets. Specifically, we used the SimSiam algorithm to pre-train the in-domain knowledge of remote sensing datasets and then transferred the obtained weights to the other scene classification datasets. Thus, we have obtained state-of-the-art results on five land cover classification datasets with varying numbers of classes and spatial resolutions. In addition, by conducting appropriate experiments, including feature pre-training using datasets with different attributes, we have identified the most influential factors that make a dataset a good choice for obtaining in-domain features. We have transferred the features obtained by pre-training SimSiam on remote sensing datasets to various downstream tasks and used them as initial weights for fine-tuning. Moreover, we have linearly evaluated the obtained representations in cases where the number of samples per class is limited. Our experiments have demonstrated that using a higher-resolution dataset during the self-supervised pre-training stage results in learning more discriminative and general representations.

An optimised deep learning approach for alzheimer’s disease classification

<p>Alzheimer’s disease (AD) is a progressive and incurable brain disorder. It starts out subtly and gets worse with time. 60 to 70 percent of dementia cases are brought on by this illness. An Alzheimer’s patient is diagnosed every two seconds, according to research. The complexity of the brain makes it often very challenging to identify in elderly people. In the area of medical imaging, deep learning is growing. Several deep learning techniques that attempted to identify and categorise the magnetic resonance imaging (MRI) brain images into four stages of AD will be compared in this work. 6400 MRI brain images were extracted from a dataset and divided into training, validation, and testing datasets. In our research on twelve deep learning architectures, inceptionV3 has given the best results with 99.56% and 97.75% accuracy on train and validation, respectively, and on test data, the model has achieved an accuracy of 95.81%. We trained the models using optimised ImageNet weights, which resulted in higher accuracy across all twelve models.</p>

Weakly-supervised learning-based pathology detection and localization in 3D chest CT scans.

Recent advancements in anomaly detection have paved the way for novel radiological reading assistance tools that support the identification of findings, aimed at saving time. The clinical adoption of such applications requires a low rate of false positives while maintaining high sensitivity. In light of recent interest and development in multi pathology identification, we present a novel method, based on a recent contrastive self-supervised approach, for multiple chest-related abnormality identification including low lung density area ("LLDA"), consolidation ("CONS"), nodules ("NOD") and interstitial pattern ("IP"). Our approach alerts radiologists about abnormal regions within a computed tomography (CT) scan by providing 3Dlocalization. We introduce a new method for the classification and localization of multiple chest pathologies in 3D Chest CT scans. Our goal is to distinguish four common chest-related abnormalities: "LLDA", "CONS", "NOD", "IP" and "NORMAL". This method is based on a 3D patch-based classifier with a Resnet backbone encoder pretrained leveraging recent contrastive self supervised approach and a fine-tuned classification head. We leverage the SimCLR contrastive framework for pretraining on an unannotated dataset of randomly selected patches and we then fine-tune it on a labeled dataset. During inference, this classifier generates probability maps for each abnormality across the CT volume, which are aggregated to produce a multi-label patient-level prediction. We compare different training strategies, including random initialization, ImageNet weight initialization, frozen SimCLR pretrained weights and fine-tuned SimCLR pretrained weights. Each training strategy is evaluated on a validation set for hyperparameter selection and tested on a test set. Additionally, we explore the fine-tuned SimCLR pretrained classifier for 3D pathology localization and conduct qualitativeevaluation. Validated on 111 chest scans for hyperparameter selection and subsequently tested on 251 chest scans with multi-abnormalities, our method achieves an AUROC of 0.931 (95% confidence interval [CI]: [0.9034, 0.9557], -value < 0.001) and 0.963 (95% CI: [0.952, 0.976], -value < 0.001) in the multi-label and binary (i.e., normal versus abnormal) settings, respectively. Notably, our method surpasses the area under the receiver operating characteristic (AUROC) threshold of 0.9 for two abnormalities: IP (0.974) and LLDA (0.952), while achieving values of 0.853 and 0.791 for NOD and CONS, respectively. Furthermore, our results highlight the superiority of incorporating contrastive pretraining within the patch classifier, outperforming Imagenet pretraining weights and non-pretrained counterparts with uninitialized weights (F1 score = 0.943, 0.792, and 0.677 respectively). Qualitatively, the method achieved a satisfactory 88.8% completeness rate in localization and maintained an 88.3% accuracy rate against falsepositives. The proposed method integrates self-supervised learning algorithms for pretraining, utilizes a patch-based approach for 3D pathology localization and develops an aggregation method for multi-label prediction at patient-level. It shows promise in efficiently detecting and localizing multiple anomalies within a singlescan.

Deep learning model for detection of hotspots using infrared thermographic images of electrical installations

Hotspots in electrical power equipment or installations are a major issue whenever it occurs within the power system. Factors responsible for this phenomenon are many, sometimes inter-related and other times they are isolated. Electrical hotspots caused by poor connections are common. Deep learning models have become popular for diagnosing anomalies in physical and biological systems, by the instrumentality of feature extraction of images in convolutional neural networks. In this work, a VGG-16 deep neural network model is applied for identifying electrical hotspots by means of transfer learning. This model was achieved by first augmenting the acquired infrared thermographic images, using the pre-trained ImageNet weights of the VGG-16 algorithm with additional global average pooling in place of conventional fully connected layers and a softmax layer at the output. With the categorical cross-entropy loss function, the model was implemented using the Adam optimizer at learning rate of 0.0001 as well as some variants of the Adam optimization algorithm. On evaluation, with a test IRT image dataset, and a comparison with similar works, the research showed that a better accuracy of 99.98% in identification of electrical hotspots was achieved. The model shows good score in performance metrics like accuracy, precision, recall, and F1-score. The obtained results proved the potential of deep learning using computer vision parameters for infrared thermographic identification of electrical hotspots in power system installations. Also, there is need for careful selection of the IR sensor’s thermal range during image acquisition, and suitable choice of color palette would make for easy hotspot isolation, reduce the pixel to pixel temperature differential across any of the images, and easily highlight the critical region of interest with high pixel values. However, it makes edge detection difficult for human visual perception which computer vision-based deep learning model could overcome.

Automatic Diabetic Retinopathy Detection Using Resnet50 and Inceptionv3

Diabetic retinopathy (DR) is an eye disease in diabetic patients and is the cause of blindness in the population. Because of the complexity of color fundus images DR classification by humans is challenging and is also an error prone task. Thus, this paper proposes automated DR classification to detect diabetic retinopathy in advance. Traditional machine learning algorithms like Logistic regression (60%) and Support Vector Machines (70%) have been employed to detect the severity of DR using information from retina images in this work. Since machine learning Algorithms provide low accuracy, deep learning algorithms have been used to improve the accuracy of the model. To develop a model from scratch is a complicated task, which can be solved using another technique called transfer learning. In this work, two CNN models are trained - Resnet50 and Inception V3 with pre-trained imageNet weights. ResNet50 is used in two ways, as a classifier, fine tuning and InceptionV3 without fine tuning. The CNN model achieved a good accuracy around (85%) using InceptionV3 without fine tuning. The dataset used was provided by Kaggle which are high resolution images of the retina which are proven to show good accuracy. Since machine learning Algorithms provide low accuracy, deep learning algorithms have been used to improve the accuracy of the model. To develop a model from scratch is a complicated task, which can be solved using another technique called transfer learning. In this work, two CNN models are trained - Resnet50 and Inception V3 with pre-trained imageNet weights. ResNet50 is used in two ways, as a classifier, fine tuning and InceptionV3 without fine tuning. The CNN model achieved a good accuracy around (85%) using InceptionV3 without fine tuning. The dataset used was provided by Kaggle which are high resolution images of the retina which are proven to show good accuracy.

Deep learning for determining the difficulty of endodontic treatment: a pilot study

BackgroundTo develop and validate a deep learning model for automated assessment of endodontic case difficulty from periapical radiographs.MethodsA dataset of 1,386 periapical radiographs was compiled from two clinical sites. Two dentists and two endodontists annotated the radiographs for difficulty using the “simple assessment” criteria from the American Association of Endodontists’ case difficulty assessment form in the Endocase application. A classification task labeled cases as “easy” or “hard”, while regression predicted overall difficulty scores. Convolutional neural networks (i.e. VGG16, ResNet18, ResNet50, ResNext50, and Inception v2) were used, with a baseline model trained via transfer learning from ImageNet weights. Other models was pre-trained using self-supervised contrastive learning (i.e. BYOL, SimCLR, MoCo, and DINO) on 20,295 unlabeled dental radiographs to learn representation without manual labels. Both models were evaluated using 10-fold cross-validation, with performance compared to seven human examiners (three general dentists and four endodontists) on a hold-out test set.ResultsThe baseline VGG16 model attained 87.62% accuracy in classifying difficulty. Self-supervised pretraining did not improve performance. Regression predicted scores with ± 3.21 score error. All models outperformed human raters, with poor inter-examiner reliability.ConclusionThis pilot study demonstrated the feasibility of automated endodontic difficulty assessment via deep learning models.

Enhancing Skin Cancer Classification using Efficient Net B0-B7 through Convolutional Neural Networks and Transfer Learning with Patient-Specific Data.

Skin cancer diagnosis challenges dermatologists due to its complex visual variations across diagnostic categories. Convolutional neural networks (CNNs), specifically the Efficient Net B0-B7 series, have shown superiority in multiclass skin cancer classification. This study addresses the limitations of visual examination by presenting a tailored preprocessing pipeline designed for Efficient Net models. Leveraging transfer learning with pre-trained ImageNet weights, the research aims to enhance diagnostic accuracy in an imbalanced multiclass classification context. The study develops a specialized image preprocessing pipeline involving image scaling, dataset augmentation, and artifact removal tailored to the nuances of Efficient Net models. Using the Efficient Net B0-B7 dataset, transfer learning fine-tunes CNNs with pre-trained ImageNet weights. Rigorous evaluation employs key metrics like Precision, Recall, Accuracy, F1 Score, and Confusion Matrices to assess the impact of transfer learning and fine-tuning on each Efficient Net variant's performance in classifying diverse skin cancer categories. The research showcases the effectiveness of the tailored preprocessing pipeline for Efficient Net models. Transfer learning and fine-tuning significantly enhance the models' ability to discern diverse skin cancer categories. The evaluation of eight Efficient Net models (B0-B7) for skin cancer classification reveals distinct performance patterns across various cancer classes. While the majority class, Benign Kertosis, achieves high accuracy (>87%), challenges arise in accurately classifying Eczema classes. Melanoma, despite its minority representation (2.42% of images), attains an average accuracy of 80.51% across all models. However, suboptimal performance is observed in predicting warts molluscum (90.7%) and psoriasis (84.2%) instances, highlighting the need for targeted improvements in accurately identifying specific skin cancer types. The study on skin cancer classification utilizes EfficientNets B0-B7 with transfer learning from ImageNet weights. The pinnacle performance is observed with EfficientNet-B7, achieving a groundbreaking top-1 accuracy of 84.4% and top-5 accuracy of 97.1%. Remarkably efficient, it is 8.4 times smaller than the leading CNN. Detailed per-class classification exactitudes through Confusion Matrices affirm its proficiency, signaling the potential of EfficientNets for precise dermatological image analysis.

On the use of contrastive learning for standard-plane classification in fetal ultrasound imaging

Background:To investigate the effectiveness of contrastive learning, in particular SimClr, in reducing the need for large annotated ultrasound (US) image datasets for fetal standard plane identification. Methods:We explore SimClr advantage in the cases of both low and high inter-class variability, considering at the same time how classification performance varies according to different amounts of labels used. This evaluation is performed by exploiting contrastive learning through different training strategies. We apply both quantitative and qualitative analyses, using standard metrics (F1-score, sensitivity, and precision), Class Activation Mapping (CAM), and t-Distributed Stochastic Neighbor Embedding (t-SNE). Results:When dealing with high inter-class variability classification tasks, contrastive learning does not bring a significant advantage; whereas it results to be relevant for low inter-class variability classification, specifically when initialized with ImageNet weights. Conclusions:Contrastive learning approaches are typically used when a large number of unlabeled data is available, which is not representative of US datasets. We proved that SimClr either as pre-training with backbone initialized via ImageNet weights or used in an end-to-end dual-task may impact positively the performance over standard transfer learning approaches, under a scenario in which the dataset is small and characterized by low inter-class variability.

A Methodology for Advanced Manufacturing Defect Detection through Self-Supervised Learning on X-ray Images

In industrial quality control, especially in the field of manufacturing defect detection, deep learning plays an increasingly critical role. However, the efficacy of these advanced models is often hindered by their need for large-scale, annotated datasets. Moreover, these datasets are mainly based on RGB images, which are very different from X-ray images. Addressing this limitation, our research proposes a methodology that incorporates domain-specific self-supervised pretraining techniques using X-ray imaging to improve defect detection capabilities in manufacturing products. We employ two pretraining approaches, SimSiam and SimMIM, to refine feature extraction from manufacturing images. The pretraining stage is carried out using an industrial dataset of 27,901 unlabeled X-ray images from a manufacturing production line. We analyze the performance of the pretraining against transfer-learning-based methods in a complex defect detection scenario using a Faster R-CNN model. We conduct evaluations on both a proprietary industrial dataset and the publicly available GDXray dataset. The findings reveal that models pretrained with domain-specific X-ray images consistently outperform those initialized with ImageNet weights. Notably, Swin Transformer models show superior results in scenarios rich in labeled data, whereas CNN backbones are more effective in limited-data environments. Moreover, we underscore the enhanced ability of the models pretrained with X-ray images in detecting critical defects, crucial for ensuring safety in industrial settings. Our study offers substantial evidence of the benefits of self-supervised learning in manufacturing defect detection, providing a solid foundation for further research and practical applications in industrial quality control.

Explainable ensemble learning method for OCT detection with transfer learning.

The accuracy and interpretability of artificial intelligence (AI) are crucial for the advancement of optical coherence tomography (OCT) image detection, as it can greatly reduce the manual labor required by clinicians. By prioritizing these aspects during development and application, we can make significant progress towards streamlining the clinical workflow. In this paper, we propose an explainable ensemble approach that utilizes transfer learning to detect fundus lesion diseases through OCT imaging. Our study utilized a publicly available OCT dataset consisting of normal subjects, patients with dry age-related macular degeneration (AMD), and patients with diabetic macular edema (DME), each with 15 samples. The impact of pre-trained weights on the performance of individual networks was first compared, and then these networks were ensemble using majority soft polling. Finally, the features learned by the networks were visualized using Grad-CAM and CAM. The use of pre-trained ImageNet weights improved the performance from 68.17% to 92.89%. The ensemble model consisting of the three CNN models with pre-trained parameters loaded performed best, correctly distinguishing between AMD patients, DME patients and normal subjects 100% of the time. Visualization results showed that Grad-CAM could display the lesion area more accurately. It is demonstrated that the proposed approach could have good performance of both accuracy and interpretability in retinal OCT image detection.

Machine learning based prediction of image quality in prostate MRI using rapid localizer images.

Diagnostic performance of prostate MRI depends on high-quality imaging. Prostate MRI quality is inversely proportional to the amount of rectal gas and distention. Early detection of poor-quality MRI may enable intervention to remove gas or exam rescheduling, saving time. We developed a machine learning based quality prediction of yet-to-be acquired MRI images solely based on MRI rapid localizer sequence, which can be acquired in a few seconds. The dataset consists of 213 (147 for training and 64 for testing) prostate sagittal T2-weighted (T2W) MRI localizer images and rectal content, manually labeled by an expert radiologist. Each MRI localizer contains seven two-dimensional (2D) slices of the patient, accompanied by manual segmentations of rectum for each slice. Cascaded and end-to-end deep learning models were used to predict the quality of yet-to-be T2W, DWI, and apparent diffusion coefficient (ADC) MRI images. Predictions were compared to quality scores determined by the experts using area under the receiver operator characteristic curve and intra-class correlation coefficient. In the test set of 64 patients, optimal versus suboptimal exams occurred in 95.3% (61/64) versus 4.7% (3/64) for T2W, 90.6% (58/64) versus 9.4% (6/64) for DWI, and 89.1% (57/64) versus 10.9% (7/64) for ADC. The best performing segmentation model was 2D U-Net with ResNet-34 encoder and ImageNet weights. The best performing classifier was the radiomics based classifier. A radiomics based classifier applied to localizer images achieves accurate diagnosis of subsequent image quality for T2W, DWI, and ADC prostate MRI sequences.

Generalisation challenges in deep learning models for medical imagery: insights from external validation of COVID-19 classifiers

The generalisability of deep neural network classifiers is emerging as one of the most important challenges of our time. The recent COVID-19 pandemic led to a surge of deep learning publications that proposed novel models for the detection of COVID-19 from chest x-rays (CXRs). However, despite the many outstanding metrics reported, such models have failed to achieve widespread adoption into clinical settings. The significant risk of real-world generalisation failure has repeatedly been cited as one of the most critical concerns, and is a concern that extends into general medical image modelling. In this study, we propose a new dataset protocol and, using this, perform a thorough cross-dataset evaluation of deep neural networks when trained on a small COVID-19 dataset, comparable to those used extensively in recent literature. This allows us to quantify the degree to which these models can generalise when trained on challenging, limited medical datasets. We also introduce a novel occlusion evaluation to quantify model reliance on shortcut features. Our results indicate that models initialised with ImageNet weights then fine-tuned on small COVID-19 datasets, a standard approach in the literature, facilitate the learning of shortcut features, resulting in unreliable, poorly generalising models. In contrast, pre-training on related CXR imagery can stabilise cross-dataset performance. The CXR pre-trained models demonstrated a significantly smaller generalisation drop and reduced feature dependence outwith the lung region, as indicated by our occlusion test. This paper demonstrates the challenging problem of model generalisation, and the need for further research on developing techniques that will produce reliable, generalisable models when learning with limited datasets.

A Modified Xception Deep Learning Model for Automatic Sorting of Olives Based on Ripening Stages

Olive fruits at different ripening stages give rise to various table olive products and oil qualities. Therefore, developing an efficient method for recognizing and sorting olive fruits based on their ripening stages can greatly facilitate post-harvest processing. This study introduces an automatic computer vision system that utilizes deep learning technology to classify the ‘Roghani’ Iranian olive cultivar into five ripening stages using color images. The developed model employs convolutional neural networks (CNN) and transfer learning based on the Xception architecture and ImageNet weights as the base network. The model was modified by adding some well-known CNN layers to the last layer. To minimize overfitting and enhance model generality, data augmentation techniques were employed. By considering different optimizers and two image sizes, four final candidate models were generated. These models were then compared in terms of loss and accuracy on the test dataset, classification performance (classification report and confusion matrix), and generality. All four candidates exhibited high accuracies ranging from 86.93% to 93.46% and comparable classification performance. In all models, at least one class was recognized with 100% accuracy. However, by taking into account the risk of overfitting in addition to the network stability, two models were discarded. Finally, a model with an image size of 224 × 224 and an SGD optimizer, which had a loss of 1.23 and an accuracy of 86.93%, was selected as the preferred option. The results of this study offer robust tools for automatic olive sorting systems, simplifying the differentiation of olives at various ripening levels for different post-harvest products.

DeepGraFT: A novel semantic segmentation auxiliary ROI-based deep learning framework for effective fundus tessellation classification

Fundus tessellation (FT) is a prevalent clinical feature associated with myopia and has implications in the development of myopic maculopathy, which causes irreversible visual impairment. Accurate classification of FT in color fundus photo can help predict the disease progression and prognosis. However, the lack of precise detection and classification tools has created an unmet medical need, underscoring the importance of exploring the clinical utility of FT. Thus, to address this gap, we introduce an automatic FT grading system (called DeepGraFT) using classification-and-segmentation co-decision models by deep learning. ConvNeXt, utilizing transfer learning from pretrained ImageNet weights, was employed for the classification algorithm, aligning with a region of interest based on the ETDRS grading system to boost performance. A segmentation model was developed to detect FT exits, complementing the classification for improved grading accuracy. The training set of DeepGraFT was from our in-house cohort (MAGIC), and the validation sets consisted of the rest part of in-house cohort and an independent public cohort (UK Biobank). DeepGraFT demonstrated a high performance in the training stage and achieved an impressive accuracy in validation phase (in-house cohort: 86.85 %; public cohort: 81.50 %). Furthermore, our findings demonstrated that DeepGraFT surpasses machine learning-based classification models in FT classification, achieving a 5.57 % increase in accuracy. Ablation analysis revealed that the introduced modules significantly enhanced classification effectiveness and elevated accuracy from 79.85 % to 86.85 %. Further analysis using the results provided by DeepGraFT unveiled a significant negative association between FT and spherical equivalent (SE) in the UK Biobank cohort. In conclusion, DeepGraFT accentuates potential benefits of the deep learning model in automating the grading of FT and allows for potential utility as a clinical-decision support tool for predicting progression of pathological myopia.

Transferability exploration of weight based on ImageNet for galaxy classification

The accurate classification of galaxies is essential for understanding the structure of the cosmos and establishing connections with host haloes. Manual classification methods are time-consuming, subjective, and prone to errors. This study investigates the efficiency of automated galaxy categorization utilising transfer learning with pre-trained models from ImageNet by utilising deep learning techniques. The Galaxy10 DECals Dataset is used in the study, and the DenseNet-121 network architecture is used for transfer learning. Experiments are used to assess the effects of various ImageNet weight configurations on the performance of the model. The findings reveal that the model trained with 20% ImageNet data achieves the highest classification accuracy of 79% for galaxies. Grad-CAM visualisation further highlights the impact of weight initialization on categorisation by displaying the different focal points of models trained with various ImageNet weights. The results indicate that using higher weights from ImageNet transfers advanced features, whereas using fewer weights transfers basic features, potentially producing inaccurate results. Prior to transfer learning, the study emphasises the significance of choosing the best weights from ImageNet. The knowledge acquired aids in the classification of galaxies and offers direction for further study.

The investigation of transferability related to the ImageNet-based transfer learning model in galaxy classification

The phenomenon of transfer learning, specifically the transferability of ImageNet, in the context of galaxy datasets has been relatively under-researched. This study seeks to address this gap by employing the Galaxy10 DECals dataset, which is a 10-classification dataset. Three classic models, namely MobileNet, VGG, and ResNet, were developed and customized to the current dataset by modifying the number of neurons in the last layer. The experimental phase is divided into three parts, including the comparison between the use of ImageNet and non-ImageNet, model performance comparison, and confusion matrix analysis. The results demonstrate that the utilization of ImageNet produces better outcomes, with the MobileNet model exhibiting the highest performance. Further analysis revealed that the inclusion of ImageNet weights can enhance the classification accuracy of some data. Although the present study was successful in achieving its objectives, future research should focus on exploring and elucidating the underlying mechanisms driving these findings.