A Light Weight Neural Network Model for Classification of Dementia

In extremely dark conditions, low-light imaging may offer spectators a rich visual experience, which is important for both military and civic applications. However, the images taken in ultra-micro light environments usually have inherent defects such as extremely low brightness and contrast, a high noise level, and serious loss of scene details and colors, which leads to great challenges in the research of low-light image and object detection and classification. The low-light night vision image used as the study object in this work has an excessively dim overall picture and very little information about the screen's features. Three algorithms, HE, AHE, and CLAHE, were used to enhance and highlight the image. The effectiveness of these image enhancement methods is evaluated using metrics such as the peak signal-to-noise ratio and mean square error, and CLAHE was selected after comparison. The target image includes vehicles, people, license plates, and objects. The gray-level co-occurrence matrix (GLCM) was used to extract the texture features of the enhanced images, and the extracted image texture features were used as input to construct a backpropagation (BP) neural network classification model. Then, low-light image classification models were developed based on VGG16 and ResNet50 convolutional neural networks combined with low-light image enhancement algorithms. The experimental results show that the overall classification accuracy of the VGG16 convolutional neural network model is 92.1%. Compared with the BP and ResNet50 neural network models, the classification accuracy was increased by 4.5% and 2.3%, respectively, demonstrating its effectiveness in classifying low-light night vision targets.

(1) Background: The purpose of this study was to investigate whether the loosening of total knee arthroplasty (TKA) implants could be detected accurately on plain radiographs using a deep convolution neural network (CNN). (2) Methods: We analyzed data for 100 patients who underwent revision TKA due to prosthetic loosening at a single institution from 2012 to 2020. We extracted 100 patients who underwent primary TKA without loosening through a propensity score, matching for age, gender, body mass index, operation side, and American Society of Anesthesiologists class. Transfer learning was used to prepare a detection model using a pre-trained Visual Geometry Group (VGG) 19. For transfer learning, two methods were used. First, the fully connected layer was removed, and a new fully connected layer was added to construct a new model. The convolutional layer was frozen without training, and only the fully connected layer was trained (transfer learning model 1). Second, a new model was constructed by adding a fully connected layer and varying the range of freezing for the convolutional layer (transfer learning model 2). (3) Results: The transfer learning model 1 gradually increased in accuracy and ultimately reached 87.5%. After processing through the confusion matrix, the sensitivity was 90% and the specificity was 100%. Transfer learning model 2, which was trained on the convolutional layer, gradually increased in accuracy and ultimately reached 97.5%, which represented a better improvement than for model 1. Processing through the confusion matrix affirmed that the sensitivity was 100% and the specificity was 97.5%. (4) Conclusions: The CNN algorithm, through transfer learning, shows high accuracy for detecting the loosening of TKA implants on plain radiographs.

In this paper, we propose a new end-to-end deep neural network model for time-series classification (TSC) with emphasis on both the accuracy and the interpretation. The proposed model contains a convolutional network component to extract high-level features and a recurrent network component to enhance the modeling of the temporal characteristics of TS data. In addition, a feedforward fully connected network with the sparse group lasso (SGL) regularization is used to generate the final classification. The proposed architecture not only achieves satisfying classification accuracy, but also obtains good interpretability through the SGL regularization. All these networks are connected and jointly trained in an end-to-end framework, and it can be generally applied to TSC tasks across different domains without the efforts of feature engineering. Our experiments in various TS data sets show that the proposed model outperforms the traditional convolutional neural network model for the classification accuracy, and also demonstrate how the SGL contributes to a better model interpretation.

In this paper, we investigate the application of quantum and quantum-inspired machine learning algorithms to stock return predictions. Specifically, we evaluate the performance of quantum neural network, an algorithm suited for noisy intermediate-scale quantum computers, and tensor network, a quantum-inspired machine learning algorithm, against classical models such as linear regression and neural networks. To evaluate their abilities, we construct portfolios based on their predictions and measure investment performances. The empirical study on the Japanese stock market shows the tensor network model achieves superior performance compared to classical benchmark models, including linear and neural network models. Though the quantum neural network model attains the lowered risk-adjusted excess return than the classical neural network models over the whole period, both the quantum neural network and tensor network models have superior performances in the latest market environment, which suggests capability of model’s capturing non-linearity between input features.

RF connectivity is pervasive in many systems to- day and can underpin fundamental services. Intentional Global Navigation Satellite System (GNSS) jamming activities are increasing across the globe, causing significant threats to real life applications from power distribution to finance and even 5G performance. The first step towards its mitigation is the detection and classification of the signal. Classification could inform an attribution picture. For example, connecting a perpetrator through the jamming signal type from a device found in their possession. This paper introduces a novel approach which utilises transfer learning from the imagery domain and considers the jamming signal power spectral density (PSD), spectrogram, raw constellation, and histogram signal representations as images. Collecting datasets large enough to train a neural network from scratch is a common problem. The use of Transfer Learning overcomes this issue. Transfer learning is applied through feature extraction using a Convolutional Neural Network (CNN) VGG16 pretrained on the ImageNet dataset. Various Machine Learning classifiers are evaluated including Support Vector Machine (SVM), Logistic Regression (LR), and Random Forest (RF). To date, prior research in this field has concentrated on spectrogram representation but our evidence shows that the novel concatenation of signal representations (PSD, spectrogram, raw constellation and histogram) is more effective, allowing the CNN to benefit from the strengths of each individual representation. The image concatenation dataset produced 98% (+/- 0.5%) classification accuracy with LR and SVM models and 96.3% (+/- 0.6%) with RF. The results, validated through 10-fold cross validation, showed that transfer learning using CNN VGG16 in conjunction with ML models LR, SVM, and RF and the concatenation of signal representations, produces high accuracy for the classification of GNSS jamming signals and outperforms previous work in the field.

Abstract:Fire detection using convolutional neural network(CNN) algorithm and surveillance cameras is a field that aims to use technology to detect and intervene in fire incidents quickly and effectively. Fires are considered one of the most dangerous disasters that can occur in buildings and facilities, so developing early fire detection systems is vital to preserve lives and property. Surveillance cameras are used to collect real-time images and videos and send them to a fire analysis and detection system. . In the event of a fire being detected, an immediate alert will be issued to the competent authorities or building owners to take the necessary measures. Develop a fire detection system using by CNN-based algorithm. This system must be accurate and cost-effective. It has many advantages use of visibility infrastructurecompared to other existing systems. There are three types. First: There is no need to update the stove structure, provided that the place is equipped with surveillance cameras that monitor fire situations and cover the entire place. Second: camera-based systems provide the actual location, that is, a complete map of the fire location, which is good , helps in detecting the fire. Third: the methods used can be highly applicable Watching fires in public places.Our system achieved excellent results with average prediction accuracy of 98.14% and 98% on the Forest dataset and the Local dataset, respectively. AlexNet is well known as a transfer learning model, where knowledge is learned by training a large amount of datasets . AlexNet is a deeper and broader CNN model introduced in 2012. AlexNet is primarily design , Alex Net consists of five convolutional layers and three FC layers. The Alex net structure is shown. The first convolutional layer implemented convolution and max pooling using local response normalization (LRN)Finally, two FC layers are used with dropout followed by a SoftMax layer following the first two convolutional layers Overlapping Max Pooling layers The third, fourth and fifth convolutional layers are directly connected. The fifth convolutional layer is followed by the Overlapping Max Pooling layer, whose output is transmitted to a series of two FC layers. The second fully connected layer feeds into a softmax classifie After all the convolution and FC layers, ReLU nonlinearity is applied. The ReLU nonlinearity of the first and second convolution layers follows a local regularization step before pooling

ABSTRACTObjectivesBenign fibro‐osseous lesions (BFOL) constitute a group of pathologic entities with marked overlapping histopathologic features but are diverse in nature and clinical behaviors. Accurate diagnoses of BFOLs necessitate clinical‐pathological correlations, which are paramount for their appropriate management. Recent research indicates the potential utility of artificial intelligence in diagnostic pathology. Here, we aimed to assess the performance of the deep convolutional neural network (DCNN) models for BFOL classification and investigate its impact on the diagnostic performance of oral pathologists.Material and MethodsMicroscopic slides from 68 patients diagnosed with cemento‐ossifying fibroma (COF), fibrous dysplasia (FD), and cemento‐osseous dysplasia (COD) were collected. The image patches from each slide were processed, augmented, and used to train and validate the five pre‐trained DCNN models for BFOL classification. The best‐performing model was selected to evaluate its diagnostic performance on the testing data set, compared with experienced oral pathologists.ResultsThe InceptionV3 model showed the highest and most balanced overall performance in BFOL classification. It demonstrated the highest accuracy (96.7%) in classifying COF, followed by COD (83.3%), and FD (80.0%), respectively. The model accuracy in identifying COF was greater than the average performance of pathologists (90.0%). However, pathologists performed better in classifying COD (87.2%) and FD (95.0%). With DCNN assistance, pathologists significantly improved the accuracy, sensitivity, and specificity in distinguishing BFOLs while reducing the average diagnosis time.ConclusionsThe DCNN model has the potential to be developed as an auxiliary tool, assisting pathologists in diagnosing BFOLs. Through ongoing refinements, artificial intelligence assistance can aid pathologists in enhancing the accuracy and efficiency of BFOL diagnosis.

This paper highlights the ability of residual convolutional neural network (ResNet) at classifying railway shelling defect dataset without the requirement for handcrafted features. A 41-convolutional layers ResNet with the residual learning block is introduced. Bottleneck architecture of three convolutional layers is used in ResNet to decrease the computation cost. VGG convolutional neural network (VGGNet) classifier and Support Vector Machine (SVM) classifiers based on Histogram of Oriented Gradient (HOG), Local Binary Patterns (LBP) and Scale-invariant Feature Transform (SIFT) are compared with ResNet classifier at classifying our dataset. The performance of ResNet presented in this paper achieves 95% TOP-1 accuracy in testing dataset. In comparison, the result of ResNet is better than VGGNet with 92% TOP-1 accuracy and HOG (42.88%), LBP (52.26%), SIFT (60.69%). In the context of designing neural computing models for ResNet analysis, this paper shows that ResNet used in our experiment is able to not only achieve the high accuracy in railway shelling defect testing dataset, but also has a faster testing speed than the other classifiers.

The study aims to compare the performance of VGG16 and VGG19 architectures in classifying rice plant disease images using the Convolutional Neural Network (CNN) method. The models were trained using Adam and SGD optimizers with various learning rates, and evaluated using accuracy, precision, recall, f1-score, and confusion matrix metrics. The results show that VGG19 with Adam's optimizer and learning rate 0.001 provides the best performance with a testing accuracy of 97.90% and the lowest validation loss value of 0.2910. VGG16 also showed good results with the highest validation accuracy of 89%, but the results were below VGG19. Based on the confusion matrix, both models accurately recognize some classes, although there are still errors in certain classes. Thus, the results obtained show that VGG19 is superior in terms of training accuracy and stability. The use of data augmentation techniques is expected to reduce the potential for overfitting.

SummaryHigh latency and power consumption are two major problems that need to be addressed in convolutional neural networks (CNN). In this paper, the convolutional layer is replaced with a discrete‐time cellular neural network (CellNN) to overcome these problems. Multiple configurations of CellNNs are trained in a framework called TensorFlow to classify objects from the CIFAR‐10 database. Effects of the number of iterations, the number of channels, batch normalization, and activation functions on the classification accuracies are presented. It is shown that TensorFlow is a tool that is capable of training discrete‐time CellNNs. Although the accuracies of the proposed networks on CIFAR‐10 are slightly lesser than the existing CNNs, with reduced parameters and multiply‐accumulates (MACs), power consumption and computation time of our networks will be less than CNNs.

An automatic plant species identification system could help to identify plant species very easily. Deep learning is an AI function which works like the human brain, with artificial neural networks. In neural networks, neuron nodes are connected like a web and used to extract higher-level features from input. Convolutional neural network (CNN), deep belief network (DBF) and recurrent neural network (RNN) etc. are deep learning networks and can extract more detailed information compared to conventional machine learning techniques [1, 10]. CNN is a very good choice for image processing, and it can work with large datasets efficiently. In our project VGG16 CNN is used to extract the features from leaf images of a simple and compound leaf. For the identification of plant species with simple and compound leaves with real complex background images, this paper proposes a fusion CNN model using original whole leaf images and patch images. A transfer learned VGG16 CNN was used for feature extraction and classification of real complex background images [1]. Feature extraction and classification are done with original (model1) and patch (model2) images separately with VGG16 CNN models. The feature maps from intermediate levels of model1 and model2 are taken, then concatenated and classified using SVM and KNN. The CNN-SVM model has the best performance over model1, model2 and CNN-KNN model. The proposed fusion model shows the efficiency of 98.6% accuracy in model evaluation and 90% accuracy in plant identification using complex background leaf images.

In this study, we used panoramic X-ray images to classify and clarify the accuracy of different dental implant brands via deep convolutional neural networks (CNNs) with transfer-learning strategies. For objective labeling, 8859 implant images of 11 implant systems were used from digital panoramic radiographs obtained from patients who underwent dental implant treatment at Kagawa Prefectural Central Hospital, Japan, between 2005 and 2019. Five deep CNN models (specifically, a basic CNN with three convolutional layers, VGG16 and VGG19 transfer-learning models, and finely tuned VGG16 and VGG19) were evaluated for implant classification. Among the five models, the finely tuned VGG16 model exhibited the highest implant classification performance. The finely tuned VGG19 was second best, followed by the normal transfer-learning VGG16. We confirmed that the finely tuned VGG16 and VGG19 CNNs could accurately classify dental implant systems from 11 types of panoramic X-ray images.

Today, detecting waste, collecting it, processing it, and getting rid of it are among the most significant environmental issues in developing and undeveloped counties. It has been observed that a large amount of garbage remains strewn on the roadside. This study presented a garbage detection technology such as machine learning and gadgets connected to the Internet of Things (IoT), such as an IP-enabled CCTV camera, to take pictures and send them to the city’s main server. The input images are transformed into a two-dimension array of integers using Python modules and divided into the garbage and no garbage classes. There is an 80:20 split between the training and testing datasets from the input dataset. Preprocessed images are then utilised as inputs for a wide range of machine learning and neural network models for classification; these include K-Nearest Neighbour (KNN), Logistic Regression (LR), Naive Bayes (NB), and Support Vector Machine (SVM). The test data sets are applied, and a confusion matrix is formed for all models to analyse the efficiency and performance of the trained models. Results from the confusion matrix are contrasted with those from the area under the Receiver characteristics operating curve (AUC). As a result, the ConvNet model is best suited for classifying garbage or no garbage present in open space, and the LR model proposed best suits the garbage detection problem. The proposed models are best suitable for improving the efficiency of existing garbage identification systems and developing a new system for smart cities.

Inspired by the information processing mechanism of the human brain, the artificial neural network (ANN) is a classic data mining method and a powerful soft computing technique. The ANN provides a valuable tool for information processing and pattern recognition, thanks to its advantages in distributed storage, parallel processing, fast problem-solving and adaptive learning. The constructive neural network (CNN) is a popular emerging neural network model suitable for processing largescale data. In this paper, a novel neural network classification model was established based on the covering algorithm (CA) and the immune clustering algorithm (ICA). The CA is highly comprehensible, and enjoys fast computing speed, and high recognition rate. However, the learning effect of this algorithm is rather poor, because the training set is randomly selected from the original data, and the number of nodes (covering number) and area being covered are greatly affected by the learning sequence. To solve the problem, the ICA was introduced to preprocess the data samples, and calculate the cluster centers based on the antibody-antigen affinity. The CA and the ICA work together to determine the covering center and radius automatically, and convert them into the weights and thresholds of the hidden layer of neural network. The number of hidden layer neurons equals the number of covering. In addition, the McCulloch-Pitts (M-P) neurons were adopted for the output layer. Based on the input feature of the hidden layer, the output feature completes the mapping from input to output, creating the final classes of the original data. The introduction of the ICA fully solves the defect of the CA. Finally, our neural network classification model was verified through experiments on real-world datasets.

Combining neuroimaging technologies and deep networks has gained considerable attention over the last few years. Instead of training deep networks from scratch, transfer learning methods have allowed retraining deep networks, which were already trained on massive data repositories, using a smaller dataset from a new application domain, and have demonstrated high performance in several application areas. In the context of a diagnosis of neurodegenerative disorders, this approach can potentially lessen the dependence of the training process on large neuroimaging datasets, and reduce the length of the training, validation, and testing process on a new dataset. To this end, the paper investigates transfer learning of deep networks, which were trained on ImageNet data, for the diagnosis of dementia. The designed networks are modifications of the AlexNet and VGG16 Convolutional Neural Networks (CNNs) and are retrained to classify Mild Cognitive Impairment (MCI), Alzheimer’s disease (AD) and normal patients using Diffusion Tensor Imaging (DTI) and Magnetic Resonance Imaging (MRI) data. An empirical evaluation using DTI and MRI data from the ADNI database supports the potential of transfer learning methods in the detection of early degenerative changes in the brain. Diagnosis of AD was achieved with an accuracy of 99.75% and a 0.995 Matthews correlation coefficient (MCC) score using transfer learning of VGG models retrained on DTI scans. Early cognitive decline was predicted with an accuracy of 93.88% and an MCC equal to 0.8602 by VGG models processing MRI data. The proposed models can be used as additional tools to support a quick and efficient diagnosis of MCI, AD and other neurodegenerative disorders.