Pre-trained Convolutional Neural Network Architecture Research Articles

Development of a facial recognition model using optimized convolution neural network with Tiny-ML and fingerprint authentication for safe lock entry

Safe boxes have been used for centuries to protect valuable items, and their design and construction have evolved over time to improve security. In this study, we carried out the design and implementation of a facial recognition technology using optimized convolutional neural networks (CNNs) with TinyML and fingerprint. The facial recognition system is designed using pre-trained CNN architectures then optimized with TinyML to make it fit into resource constrained edge device, the fingerprint recognition is designed using ridge matching biometric algorithm which fits in perfectly into the microcontroller-based system. The safe box was fabricated using sheet metal and insulated using heat resistive material (fiber glass), components like OV2640 camera module, ESP32-S3, R305 Fingerprint sensor, 4.3 inches touch screen TFT display and a charging module to manage and charge the systems battery were used. The system was implemented to be a level two security and two-way verification system using secured Wi-Fi to access the system before the facial and fingerprint recognition functionality is activated, the sequence of operation can either be facial recognition before fingerprint recognition or fingerprint recognition before facial recognition. Evaluating the system’s facial recognition performance with confusion matrix, of all the optimized CNN architectures like VGGNet, SENet, RestNet, MobileNet-V2, EfficientNet-B0, SENet Architecture performed best outputting an accuracy of 98%, precision of 92%, also SENet unoptimized model performed same, MobileNet-V2 performed optimally in the inference speed test with a speed of 15ms/images with the fingerprint also evaluated using confusion matrix and perform with 98% accuracy. It was noticed that the inference speed of the unoptimized model is faster than that of the TinyML optimized model which is justifiable based on the processing clock speed, memory usage, GPU and architecture of the microcontroller/microprocessor. From the result obtained TinyML proves to be suitable mean for edge AI (artificial intelligence) and is recommended for future research and development.

Deep Learning based Smart Image Search Engine

Abstract: This paper introduces a new reverse search engine integration into content-based image retrieval (CBIR) systems that employs convolutional neural networks (CNNs) for feature extraction. It generates global descriptors using pre-trained CNN architectures such as ResNet50, InceptionV3, and InceptionResNetV2. It retrieves visually similar images without depending on linguistic annotations. Comparative analysis against existing methods, such as Gabor Wavelet, CNN-SVM, Metaheuristic Algorithm, etc., has been tested, and it proves the superiority of the proposed algorithm, the Cartoon Texture Algorithm, in CBIR. As the Internet sees an exponential growth of different data types, the importance of CBIR continues to grow. In order to efficiently retrieve images, solely relying on image features while ignoring metadata is exactly what we need. As such, this paper is a reminder of the need for CBIR in this changing world. They showed that CBIR continues to be quite effective in the age of the Internet. Their proposed model for CBIR, which integrates ResNet-50-based feature extraction, a neural network model trained on different image datasets, and clustering techniques to make retrieval fast, provides a significant improvement in accuracy and efficiency for content-dependent image retrieval. This methodology is likely to be very useful as we work with the increasingly huge data of vision and beyond on the Internet. It provides a good basis for an effective image search and retrieval system

Enhancing building segmentation by deep multiview classification for advancing sustainable urban development

Accurate building segmentation plays a crucial role in a wide range of applications such as urban planning, monitoring, and mapping. Different deep learning models were employed for building segmentation. However, these models analyze images from a single view. Given the limitations of single-view building segmentation models, our research aims to enhance accuracy by proposing a novel multi-view U-Net deep model for accurate building segmentation that incorporates multiple views of the images. We employ two pre-trained convolutional neural network architectures, MobileNetV2 and ResNet50, to extract features representing two different views of our images. By fusing these features, our proposed method effectively captures complementary information, leading to enhanced segmentation accuracy. To further improve the model’s performance, we incorporate skip connections and up-convolutional layers to ensure fine-grained feature propagation. Our experimental results on a large building dataset demonstrate a significant improvement in segmentation accuracy 91% compared to state-of-the-art methods, highlighting the effectiveness of our multiview fusion approach. The experimental results enhance the benefits of creating different views by adopting the novel concept proposed in this paper. This research has the potential to redefine the landscape of building segmentation in applications such as urban planning and mapping. We also conducted a test on a large study area (city scale of Belval–Luxembourg). This demonstrates the capabilities of our method and its efficiency in segmenting satellite images from a large extent area and reinforces its potential for real-world applications.

An analysis of Heptoprotective effects of plant extract on albino rats by Deep Learning analysis using ResNet and VGG16 algorithms

The hepatoprotective and anti-inflammatory effects of Terminalia catappa L. and Pergularia daemia hydroalcoholic extracts in albino rats with CCl4-induced liver damage (HAETC and HAEPD, respectively). The extracts significantly reduced the harm induced by CCl4 when given orally at a dosage of 200mg/kg. They also showed strong hepatoprotective and antioxidant properties. Wistar albino rats in good health were divided into seven groups, each with six animals. Group I acted as the control and received no treatment, whereas Group II was given 30% CCl4 (1ml/kg, intraperitoneally) to cause hepatotoxicity. After administering Silymarin (25 mg/kg, orally) as per protocol, Group III underwent CCl4 therapy. The CCl4-induced rats showed significant reductions in antioxidant enzymes like SOD, catalase, glutathione peroxidase (GPX), and glutathione-S-transferase (GST), as well as increases in serum marker enzymes like SGOT, SGPT, alkaline phosphatase (ALK), ACP, LDH, and lipid peroxidation (LPO) activity. However, rats given HAEPD and HAETC showed significant improvements, returning these parameters to levels that are almost normal (P 0.001).The hepatoprotective and antioxidant benefits of Terminalia catappa L. (HAETC) and Pergularia daemia (HAEPD) against CCl4-induced hepatotoxicity in rats were further validated by histopathological analysis of liver tissues. It encourages the use of pre-trained CNN architectures, such as VGG16 and ResNet (Residual Network), in the proposed method, which uses convolutional neural networks (CNNs) for histopathological image analysis to assess the hepatoprotective properties of Pergularia daemia and Terminalia catappa L. leaf extracts against CCl4-induced hepatotoxicity. These well-known CNN models may be modified for precise and effective histopathological image interpretation, assisting in the evaluation of the impact of plant extracts on liver health.

Boosting the performance of pretrained CNN architecture on dermoscopic pigmented skin lesion classification.

Pigmented skin lesions (PSLs) pose medical and esthetic challenges for those affected. PSLs can cause skin cancers, particularly melanoma, which can be life-threatening. Detecting and treating melanoma early can reduce mortality rates. Dermoscopic imaging offers a noninvasive and cost-effective technique for examining PSLs. However, the lack of standardized colors, image capture settings, and artifacts makes accurate analysis challenging. Computer-aided diagnosis (CAD) using deep learning models, such as convolutional neural networks (CNNs), has shown promise by automatically extracting features from medical images. Nevertheless, enhancing the CNN models' performance remains challenging, notably concerning sensitivity. In this study, we aim to enhance the classification performance of selected pretrained CNNs. We use the 2019 ISIC dataset, which presents eight disease classes. To achieve this goal, two methods are applied: resolution of the dataset imbalance challenge through augmentation and optimization of the training hyperparameters via Bayesian tuning. The performance improvement was observed for all tested pretrained CNNs. The Inception-V3 model achieved the best performance compared to similar results, with an accuracy of 96.40% and an AUC of 0.98. According to the study, classification performance was significantly enhanced by augmentation and Bayesian hyperparameter tuning.

The investigation of transferability utilizing the ImageNet weight-based pretrained model for medical image classification: A case study on kidney CT images

Due to recent growth in technology, machine learning has emerged to be an effective auxiliary tool in medical field. However, the effectiveness of transfer learning architectures trained on non-medical image data remains unclear. In this paper, two VGG-16 models, a type of pre-trained Convolutional Neural Network architecture, were constructed to classify kidney CT images that belong to four categories: normal, stone, cyst, and tumor. Two VGG-16 models have identical parameters except for the pre-trained weights: one has pre-trained weights trained on ImageNet, and the other one trained on a random large-scale dataset. To gather a more detained insight into models performances, saliency maps and Grad-CAM are employed to assess the models' ability to extract relevant features from the CT images. The result demonstrated that VGG-16 model that is trained on ImageNet can achieve 98.96% accuracy, which is about 30% higher than the other VGG-16 model. The saliency maps and Grad-CAM also support the difference in test accuracy: the model with random pre-trained dataset has saliency map that highlights the whole picture and Grad-CAM image that does not highlight any part of the CT image data, showing that it cannot correctly locate the key features. Additionally, the model with ImageNet can correctly highlight the principal features in both maps. In this study, the utilization of ImageNet is proven to be effective in the usage of transfer learning in processing medical image. Future research and exploration should focus on further enhancing the application of transfer learning in the medical field.

ResNet-Based Parkinson's Disease Classification

Parkinson&#x0027;s Disease (PD) is a brain disorder that leads to shaking, stiffness, and difficulty with walking, balance, and coordination. The symptoms usually begin gradually and get worse over time. Early diagnosis is very important, because treatments are more effective and easier to perform during the early stages of PD. However, early diagnosis is challenging, because the symptoms start gradually, and at the early stages, they are not very noticeable. In this paper, we propose a method that uses ResNet50, a Residual Network that has 50 layers, to help diagnosis PD. The data used is a collection of frequency features acquired by applying spectral analysis strategies to the speech recordings of the patient. We then convert the frequency features into a 2-dimensional heat map. This heat map is passed to ResNet50, which predicts whether the patient has PD or not. We have conducted experiments and compared the accuracy with several state-of-art methods. The results have demonstrated the feasibility and robustness of the proposed method. <i>Impact Statement</i> &#x2014; Parkinson&#x0027;s Disease (PD) is the most common movement disorder with more than 10 million PD patients worldwide. In the United States, around 60,000 are diagnosed with PD every year. Treatments are more effective at early stage, so early diagnosis is very important. A simple and effective classifier to diagnose PD patients is therefore crucial for doctors and patients. For the last decade, Convolutional Neural Networks (CNN) are the dominant approach for image classifications. Our proposed method benefits from the versatility, solid performance of pretrained CNN architecture by converting the frequency domain features of patients' speech recordings into heat maps. Our method achieved an accuracy of 90.7&#x0025; in diagnosis of PD by only using Tunable Q-Factor Wavelet Transform (TQWT) features, outperforming other state-of-art methods. Therefore, the conversion from non-graphical features to heat maps provides simple, accurate deep network for PD diagnosis that is fast to train, and only requires TQWT features.

Breast cancer classification with histopathological image based on machine learning

&lt;span lang="EN-US"&gt;Breast cancer represents one of the most common reasons for death in the worldwide. It has a substantially higher death rate than other types of cancer. Early detection can enhance the chances of receiving proper treatment and survival. In order to address this problem, this work has provided a convolutional neural network (CNN) deep learning (DL) based model on the classification that may be used to differentiate breast cancer histopathology images as benign or malignant. Besides that, five different types of pre-trained CNN architectures have been used to investigate the performance of the model to solve this problem which are the residual neural network-50 (ResNet-50), visual geometry group-19 (VGG-19), Inception-V3, and AlexNet while the ResNet-50 is also functions as a feature extractor to retrieve information from images and passed them to machine learning algorithms, in this case, a random forest (RF) and k-nearest neighbors (KNN) are employed for classification. In this paper, experiments are done using the BreakHis public dataset. As a result, the ResNet-50 network has the highest test accuracy of 97% to classify breast cancer images.&lt;/span&gt;

Comparative Analysis of Pre-Trained CNN Architectures for Apple Foliar Disease Classification

Apple foliar diseases can significantly impact crop yield and quality. Early and accurate disease detection allows for timely disease management, reducing losses. This paper presents a comparative analysis of pretrained Convolutional neural network (CNN) architectures for automated apple foliar disease classification from leaf images. Eight CNN models, including DenseNet121, DenseNet201, ResNet50V2, InceptionV3, InceptionResNetV2, MobileNet, MobileNetV2 and Xception, were evaluated using an apple leaf dataset with five disease classes - scab, rust, black rot, multiple diseases, and healthy. Models were trained using transfer learning by fine-tuning on the dataset with different hyperparameters. Comparative performance analysis across five cases with varying batch size, optimizer, learning rate and epochs was conducted. Results indicate MobileNet architecture achieved best accuracy of 97%, outperforming other models. This demonstrates the potential of using pretrained CNNs for robust apple disease classification to enable precision agriculture and improved crop health monitoring. Key Words: apple diseases, convolutional neural networks, transfer learning, image classification

Gray-to-color image conversion in the classification of breast lesions on ultrasound using pre-trained deep neural networks.

Breast ultrasound (BUS) image classification in benign and malignant classes is often based on pre-trained convolutional neural networks (CNNs) to cope with small-sized training data. Nevertheless, BUS images are single-channel gray-level images, whereas pre-trained CNNs learned from color images with red, green, and blue (RGB) components. Thus, a gray-to-color conversion method is applied to fit the BUS image to the CNN's input layer size. This paper evaluates 13 gray-to-color conversion methods proposed in the literature that follow three strategies: replicating the gray-level image to all RGB channels, decomposing the image to enhance inherent information like the lesion's texture and morphology, and learning a matching layer. Besides, we introduce an image decomposition method based on the lesion's structural information to describe its inner and outer complexity. These gray-to-color conversion methods are evaluated under the same experimental framework using a pre-trained CNN architecture named ResNet-18 and a BUS dataset with more than 3000 images. In addition, the Matthews correlation coefficient (MCC), sensitivity (SEN), and specificity (SPE) measure the classification performance. The experimental results show that decomposition methods outperform replication and learning-based methods when using information from the lesion's binary mask (obtained from a segmentation method), reaching an MCC value greater than0.70 and specificity up to0.92, although the sensitivity is about0.80. On the other hand, regarding the proposed method, the trade-off between sensitivity and specificity is better balanced, obtaining about0.88 for both indices and an MCC of0.73. This study contributes to the objective assessment of different gray-to-color conversion approaches in classifying breast lesions, revealing that mask-based decomposition methods improve classification performance. Besides, the proposed method based on structural information improves the sensitivity, obtaining more reliable classification results on malignant cases and potentially benefiting clinical practice.

Differentiation of lung and breast cancer brain metastases: Comparison of texture analysis and deep convolutional neural networks.

Metastases are the most common neoplasm in the adult brain. In order to initiate the treatment, an extensive diagnostic workup is usually required. Radiomics is a discipline aimed at transforming visual data in radiological images into reliable diagnostic information. We aimed to examine the capability of deep learning methods to classify the origin of metastatic lesions in brain MRIs and compare the deep Convolutional Neural Network (CNN) methods with image texture based features. One hundred forty three patients with 157 metastatic brain tumors were included in the study. The statistical and texture based image features were extracted from metastatic tumors after manual segmentation process. Three powerful pre-trained CNN architectures and the texture-based features on both 2D and 3D tumor images were used to differentiate lung and breast metastases. Ten-fold cross-validation was used for evaluation. Accuracy, precision, recall, and area under curve (AUC) metrics were calculated to analyze the diagnostic performance. The texture-based image features on 3D volumes achieved better discrimination results than 2D image features. The overall performance of CNN architectures with 3D inputs was higher than the texture-based features. Xception architecture, with 3D volumes as input, yielded the highest accuracy (0.85) while the AUC value was 0.84. The AUC values of VGG19 and the InceptionV3 architectures were 0.82 and 0.81, respectively. CNNs achieved superior diagnostic performance in differentiating brain metastases from lung and breast malignancies than texture-based image features. Differentiation using 3D volumes as input exhibited a higher success rate than 2D sagittal images.

A novel approach for image-based olive leaf diseases classification using a deep hybrid model

The olive tree is affected by a variety of diseases. To identify these diseases, many farmers typically use traditional methods that require a lot of effort and specialization. These methods include visually observing the tree or conducting tests in a laboratory. Fortunately, recent progress in machine learning (ML) and deep learning (DL) has demonstrated promising potential to automatically classify diseases with both high accuracy and speed. However, as indicated by the literature, only a few studies are utilizing ML and DL techniques for identifying and categorizing diseases that affect olive trees. Therefore, in this study, we collected a dataset containing 4138 images of olive leaves from various sources. The dataset comprises four categories: three representing diseases and one denoting a healthy category. We also introduced an innovative approach to classify olive leaf diseases by combining deep learning architectures, specifically convolutional neural networks (CNNs), with machine learning classifiers. In this approach, we developed a total of 30 distinct deep hybrid models (DHMs), utilizing six pre-trained convolutional neural network architectures (VGG19, ResNet50, MobileNetV2, InceptionV3, DenseNet201, and EfficientNetB0) as feature extractors, along with five machine learning classifiers (MLP, LR, RF, SVM, and DT). To assess the performance of the DHMs, we used performance evaluation metrics (Accuracy, Precision, Recall, F1-score) and we conducted an assessment to validate the reliability rating of the DHMs using a cross-validation technique. Additionally, we employed the Non-Parametric ScottKnott ESD (NPSK) test to assess the ranking of the best DHMs. The study’s findings revealed that the most efficient deep hybrid model was achieved by using the EfficientNetB0 model in combination with a logistic regression classifier, achieving an impressive accuracy score of 96.14%. Our approach has the potential to significantly assist olive farmers in rapidly and accurately identifying diseases, thereby potentially reducing economic losses.

Leukocytes Classification Using Convolutional Neural Network

Maintaining a healthy life is very important in today’s world. Our body has ability to fight against contagious diseases with the aid of white blood corpuscles (WBC) which generates natural immune system. Preserving a good WBC count is crucial as it leads to various hematological problems, one among them is Leukemia, a condition which results in blood cancer, huge accumulation of WBC cells in bone marrow, an abnormal growth of WBC cells, hinders natural immune system fighting against infectious diseases. White blood cells can be categorised into Eosinophils, Lymphocytes, Monocytes, Neutrophils and Basophils. In this study, a pre-trained Convolutional neural network architecture called Le Net is used to automatically classify WBC cells. It efficiently classifies WBC cells from the given input sample blood cell images compared to other pre trained models such as Alex-Net and custom-built CNN called white capsule net. The hyper parameters Le Net are fine tuned to yield higher accuracy of 96%.

Mobile Application for Medicinal Plants Recognition from Leaf Image Using Convolutional Neural Network

Indonesia is a country that has thousands of plant types that can be used as traditional medicine. However, some people have not utilized this potential optimally due to the lack of knowledge about medicinal plants' types, benefits, and substances. Therefore, there is a need to develop an application that can identify medicinal plants that grow in Indonesia and provide information about the benefits and content of the substances contained in them. In this study, medicinal plants will be recognized using a mobile application from leaf images based on a pre-trained convolutional neural network (CNN) with a transfer learning technique. Three pre-trained CNN architectures, namely VGG-16, MobileNetV2, and DenseNet-121, are explored for medicinal plant recognition. Hyperparameter tuning is performed at the fully connected layer of all architectures with 20 possible modifications to find the best model. The experimental results on 24 types of medicinal plants show that the model based on MobileNetV2 achieves the best classification accuracy of 97.74%. The best model is obtained by modifying the fully connected layer of MobileNetV2 into three dense layers with the number of neurons 736, 448, and 928, respectively. After the application recognizes the types of medicinal plants, information about the benefits and substances contained in them is displayed to the user.

Performance improvement and complexity reduction in the classification of EMG signals with mRMR-based CNN-KNN combined model

For real-time applications, the performance in classifying the movements should be as high as possible, and the computational complexity should be low. This paper focuses on the classification of five upper arm movements which can be provided as a control for human-machine interface (HMI) based applications. The conventional machine learning algorithms are used for classification with both time and frequency domain features, and k-nearest neighbor (KNN) outplay others. To further improve the classification accuracy, pretrained CNN architectures are employed which leads to computational complexity and memory requirements. To overcome this, the deep convolutional neural network (CNN) model is introduced with three convolutional layers. To further improve the performance which is the key idea behind real-time applications, a hybrid CNN-KNN model is proposed. Even though the performance is high, the computation costs of the hybrid method are more. Minimum redundancy maximum relevance (mRMR), a feature selection method makes an effort to reduce feature dimensions. As a result, better performance is achieved by our proposed method CNN-KNN with mRMR which reduces computational complexity and memory requirement with a mean prediction accuracy of about 99.05±0.25% with 100 features.

Deep Network Architectures as Feature Extractors for Multi-Label Classification of Remote Sensing Images

Data in the form of images are now generated at an unprecedented rate. A case in point is remote sensing images (RSI), now available in large-scale RSI archives, which have attracted a considerable amount of research on image classification within the remote sensing community. The basic task of single-target multi-class image classification considers the case where each image is assigned exactly one label from a predefined finite set of class labels. Recently, however, image annotations have become increasingly complex, with images labeled with several labels (instead of just one). In other words, the goal is to assign multiple semantic categories to an image, based on its high-level context. The corresponding machine learning tasks is called multi-label classification (MLC). The classification of RSI is currently predominantly addressed by deep neural network (DNN) approaches, especially convolutional neural networks (CNNs), which can be utilized as feature extractors as well as end-to-end methods. After only considering single-target classification for a long period, DNNs have recently emerged that address the task of MLC. On the other hand, trees and tree ensembles for MLC have a long tradition and are the best-performing class of MLC methods, but need predefined feature representations to operate on. In this work, we explore different strategies for model training based on the transfer learning paradigm, where we utilize different families of (pre-trained) CNN architectures, such as VGG, EfficientNet, and ResNet. The architectures are trained in an end-to-end manner and used in two different modes of operation, namely, as standalone models that directly perform the MLC task, and as feature extractors. In the latter case, the learned representations are used with tree ensemble methods for MLC, such as random forests and extremely randomized trees. We conduct an extensive experimental analysis of methods over several publicly available RSI datasets and evaluate their effectiveness in terms of standard MLC measures. Of these, ranking-based evaluation measures are most relevant, especially ranking loss. The results show that, for addressing the RSI-MLC task, it is favorable to use lightweight network architectures, such as EfficientNet-B2, which is the best performing end-to-end approach, as well as a feature extractor. Furthermore, in the datasets with a limited number of images, using traditional tree ensembles for MLC can yield better performance compared to end-to-end deep approaches.

Assessment of Convolutional Neural Network Pre-Trained Models for Detection and Orientation of Cracks.

Failure due to cracks is a major structural safety issue for engineering constructions. Human examination is the most common method for detecting crack failure, although it is subjective and time-consuming. Inspection of civil engineering structures must include crack detection and categorization as a key component of the process. Images can automatically be classified using convolutional neural networks (CNNs), a subtype of deep learning (DL). For image categorization, a variety of pre-trained CNN architectures are available. This study assesses seven pre-trained neural networks, including GoogLeNet, MobileNet-V2, Inception-V3, ResNet18, ResNet50, ResNet101, and ShuffleNet, for crack detection and categorization. Images are classified as diagonal crack (DC), horizontal crack (HC), uncracked (UC), and vertical crack (VC). Each architecture is trained with 32,000 images equally divided among each class. A total of 100 images from each category are used to test the trained models, and the results are compared. Inception-V3 outperforms all the other models with accuracies of 96%, 94%, 92%, and 96% for DC, HC, UC, and VC classifications, respectively. ResNet101 has the longest training time at 171 min, while ResNet18 has the lowest at 32 min. This research allows the best CNN architecture for automatic detection and orientation of cracks to be selected, based on the accuracy and time taken for the training of the model.

Performance analysis of pretrained convolutional neural network models for ophthalmological disease classification.

This study aimed to evaluate the classification performance of pretrained convolutional neural network models or architectures using fundus image dataset containing eight disease labels. A publicly available ocular disease intelligent recognition database has been used for the diagnosis of eight diseases. This ocular disease intelligent recognition database has a total of 10,000 fundus images from both eyes of 5,000 patients for the following eight diseases: healthy, diabetic retinopathy, glaucoma, cataract, age-related macular degeneration, hypertension, myopia, and others. Ocular disease classification performances were investigated by constructing three pretrained convolutional neural network architectures including VGG16, Inceptionv3, and ResNet50 models with adaptive moment optimizer. These models were implemented in Google Colab, which made the task straight-forward without spending hours installing the environment and supporting libraries. To evaluate the effectiveness of the models, the dataset was divided into 70%, 10%, and 20% for training, validation, and testing, respectively. For each classification, the training images were augmented to 10,000 fundus images. ResNet50 achieved an accuracy of 97.1%; sensitivity, 78.5%; specificity, 98.5%; and precision, 79.7%, and had the best area under the curve and final score to classify cataract (area under the curve = 0.964, final score = 0.903). By contrast, VGG16 achieved an accuracy of 96.2%; sensitivity, 56.9%; specificity, 99.2%; precision, 84.1%; area under the curve, 0.949; and final score, 0.857. These results demonstrate the ability of the pretrained convolutional neural network architectures to identify ophthalmological diseases from fundus images. ResNet50 can be a good architecture to solve problems in disease detection and classification of glaucoma, cataract, hypertension, and myopia; Inceptionv3 for age-related macular degeneration, and other disease; and VGG16 for normal and diabetic retinopathy.

Handwritten Gujarati Numerals Classification Based on Deep Convolution Neural Networks Using Transfer Learning Scenarios

In recent years, handwritten numeral classification has achieved remarkable attention in the field of computer vision. Handwritten numbers are difficult to recognize due to the different writing styles of individuals. In a multilingual country like India, negligible research attempts have been carried out for handwritten Gujarati numerals recognition using deep learning techniques compared to the other regional scripts. The Gujarati digit dataset is not available publicly and deep learning requires a large amount of labeled data for the training of the models. If the number of annotated data is not sufficient enough to train Convolutional Neural Networks (CNN) from the scratch, transfer learning can be applied. However, the issue arises by using transfer learning is that how deep to fine-tune the pre-trained convolutional neural network while training the target model. In this paper, we addressed these problems using three deep transfer learning scenarios to classify handwritten Gujarati numerals from the images of zero to nine. We presented transfer learning scenarios using ten pre-trained CNN architectures including LeNet, VGG16, InceptionV3, ResNet50, Xception, ResNet101, MobileNet, MobileNetV2, DenseNet169 and EfficientNetV2S to find the best performing model by freezing and fine-tuning the weight parameters. We implemented the pre-trained models using a self-created handwritten Gujarati digit dataset with 8000 images of zero to nine digits with data augmentation. Exhaustive experiments are performed using various performance evaluation matrices. EfficientNetV2S model showed promising results among all the models including three transfer learning scenarios and achieved 98.39% training accuracy, 97.92% testing accuracy, 97.69% f1-score, and 97.15% AUC. Our handwritten Gujarati digit dataset is available on <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/Parth-Goel/gujarati-handwritten-digit-dataset/</uri> .

Landmark Recognition Model for Smart Tourism using Lightweight Deep Learning and Linear Discriminant Analysis

Scene recognition algorithm is crucial for landmark recognition model development. Landmark recognition model is one of the main modules in the intelligent tour guide system architecture for the use of smart tourism industry. However, recognizing the tourist landmarks in the public places are challenging due to the common structure and the complexity of scene objects such as building, monuments and parks. Hence, this study proposes a super lightweight and robust landmark recognition model by using the combination of Convolutional Neural Network (CNN) and Linear Discriminant Analysis (LDA) approaches. The landmark recognition model was evaluated by using several pretrained CNN architectures for feature extraction. Then, several feature selections and machine learning algorithms were also evaluated to produce a super lightweight and robust landmark recognition model. The evaluations were performed on UMS landmark dataset and Scene-15 dataset. The results from the experiments have found that the Efficient Net (EFFNET) with CNN classifier are the best feature extraction and classifier. EFFNET-CNN achieved 100% and 94.26% classification accuracy on UMS-Scene and Scene-15 dataset respectively. Moreover, the feature dimensions created by EFFNet are more compact compared to the other features and even have significantly reduced for more than 90% by using Linear Discriminant Analysis (LDA) without jeopardizing classification performance but yet improved its performance.