Neural Networks For Image Recognition Research Articles

Enhanced Memristive Performance via a Vertically Heterointerface in Nanocomposite Thin Films for Artificial Synapses.

Memristors can be used to mimic synaptic behavior in artificial neural networks, which makes them a key component in neuromorphic computing and holds promise for advancing the field. In this study, a memory artificial synaptic device based on ZnO-BaTiO3 (ZnO-BTO) vertically aligned nanocomposite thin films was prepared. The vertical interface between the two phases can be used as a conduit for oxygen vacancy (OV) accumulation and a channel for OV movement, which greatly optimizes the resistive switching performance of the device and has the potential for multistage storage. By applying different pulse sequences to the device, the conductance of the device is adjusted from multiple angles, and a variety of synaptic functions are simulated, such as paired-pulse facilitation, spike-timing-dependent plasticity, short-term plasticity to long-term plasticity (STP-LTP), and long-term potentiation/depression (LTP/LTD). Finally, we construct a neural network for image recognition, and the recognition accuracy can reach 91%. Our study demonstrates the feasibility of using composite thin-film vertical interface to regulate the resistive performance of memristors and its great potential in artificial synaptic simulation and neuromorphic computing.

Guest Editorial: Foreword to the Special Issue on Advanced Research and Applications of Deep Learning and Neural Network in Image Recognition

Over the last two decades, the realm of image recognition has undergone a remarkable transformation, characterized by an astonishing pace of advancement [...]

Point of Sale System Using Convolutional Neural Network for Image Recognition in Grocery Store

The history of point of sale already has been told from a long time ago. The business nowadays is opting for the point-of-sale transactions because it was easy to sell the item to people face to face. This will build some trust between the cashier and the customer. The popular store that always customer need was the grocery store. However, the grocery store nowadays still not has a good feature for the point-of-sale system. The cashier still needs to scan the item through barcode scanner. This idea was led to make the point-of-sale transactions easier in the grocery store by applying the machine learning to the system. The problem for this project is the customer wait for a long time for their point-of-sale transactions to finish when bought the grocery items. The aim of this project is to detect the grocery items with convolutional neural network model for image recognition through camera within the main user interface. The Agile Development Life Cycle (ADLC) method is used in the development of Point-of-Sale System using Machine Learning for Image Recognition in Grocery Store. Moreover, this project is to evaluate the usability of the system using Post-Study System Usability Questionnaire (PSSUQ) approach. The PSSUQ evaluation is evaluated by the users of the system. The results of PSSUQ stated that the users satisfied with the system. The future research for this project is to make the point-of-sale system with a better model in the future. In conclusion, the system is works well and machine learning image recognition model also can detect the grocery item clearly.

High performance resistive random access memory based on Ag/TiO2 Nanorods/FTO for image recognition applications

This study investigates the electrical and synaptic properties of Ag/TiO2 nanorod/FTO-based resistive random access memory (RRAM) devices, focusing on the impact of different seed layer thicknesses on nanorod thickness and RRAM performance. The devices exhibit notable achievements, including a high DC endurance of up to 150,000 cycles, a self-compliance function of 10−2 A, and a resistance switching ratio exceeding 100-fold. Notably, the device with a 3.5 μm TiO2 nanorod thickness demonstrates exceptional stability. Analysis of the IV curve reveals that the switching mechanism is attributed to space-charge-limited conduction (SCLC) resulting from electron trapping in oxygen vacancy traps. Furthermore, the device maintains stable synaptic properties even after undergoing 20 cycles of long-term potentiation and depression. When employed in the context of artificial neural networks for image recognition, these devices attain high accuracy rates in recognizing handwritten digits, achieving approximately 90% accuracy for 8 × 8 pixel images and approximately 88% for 28 × 28 pixel images, demonstrating their suitability for practical applications.

THU260 Development Of Plantar Thermography For Screening Of Diabetic Neuropathy

Abstract Disclosure: V. Panamonta: None. R. Jerawatana: None. P. Ariyaprayoon: None. P. Looareesuwan: None. B. Ongphiphadhanakul: None. C. Sriphrapradang: None. B. Ongphiphadhanakul: None. Background: Diabetic foot ulcers are a major complication of diabetes. They are also the leading cause of nontraumatic lower extremity amputation. The incidence of diabetic foot ulcers is highest in those patients with poor glycemic control and neuropathy. In people with diabetic neuropathy, the feet temperature may change due to decreased blood flow and nerve damage. Thermography has been proposed as a noninvasive modality to identify patients at risk for diabetic foot ulcers. In this study, we used thermography and deep learning to stratify patients with diabetes at risk for developing a foot ulcer. Methods: We prospectively record clinical data and plantar thermogram in adult diabetic patients underwent diabetic foot screening at outpatient clinics at Ramathibodi hospital, Bangkok, Thailand during September to December 2022. Altogether, there were 245 thermal images were analyzed using a deep learning algorithm to determine the risk for diabetic foot ulcers by transfer learning using the pre-trained VGG16 model which is a convolutional neural network for image recognition. Twenty percent of the images were set aside for testing purposes while 20% of the images served as the validation set during training of the neural network. The neural network was trained and weighed more toward higher sensitivity of identifying at risk feet for screening purpose. Results: The study sample consisted of individuals with a mean age of 62.5±12.5 years, 57.1% who had diabetes for more than 10 years. The majority of the participants were females (57.1%). The average body mass index (BMI) was 27.8±9.6 (kg/m2). There were 186 thermal images classified as category 0 and 59 images classified as categories 1 to 3 according to the American Diabetes Association risk classification. The trained neural network achieved 75 % accuracy with 95 % sensitivity and 67 % specificity for classifying thermograms as higher than normal risk in the training dataset. For the testing dataset, the sensitivity and specificity was 75% and 41%, respectively. The accuracy, however, decreased to 50% due mainly to the misclassification of normal thermograms from abnormal. Conclusions: These results suggest that thermography combined with deep learning could potentially be developed for screening purpose to stratify patients at risk of developing diabetic foot ulcers. Further research is needed to validate the results in larger datasets and to explore alternative algorithms for predicting diabetic foot ulcers. Presentation: Thursday, June 15, 2023

Co-existence of digital and analog resistive switching in 2D layered BiOI nanosheets for synaptic applications

The emulation of the biological synapses is essential for brain-inspired computing which is expected to overcome the traditional von Neumann bottleneck. Thus, synaptic memristor with analog resistive switching (RS) is highly desirable in non-volatile memristors for future neuromorphic computing. Herein, the co-existence of digital and analog RS can be observed in two-dimensional (2D) layered BiOI nanosheets sandwiched by the top and bottom Pt electrodes. The vertical Pt/BiOI/Pt memristors demonstrate typical bipolar RS behaviors with a large ON/OFF ratio of 1.0 × 103 and long retention time up to 1.6 × 104 s under a relatively large operation voltage. When the operation voltages are reduced to 1 V, analog RS behaviors with a series of tunable resistance states can be observed. The adjustable resistance states can be utilized to emulate “learning-forgetting” experience of human brain. Repeatable long-term potentiation (LTP) and long-term depression (LTD) cycles can be implemented based on the synaptic memristors, which can be used for simulation of artificial neural network for image recognition with accuracy up to 91.15 %. Moreover, Pavlov’s dog experiment is successfully emulated based on the synaptic memristors. This study suggests good prospects of the synaptic memristors based on BiOI nanosheets for future neuromorphic computing.

MCA: Multidimensional collaborative attention in deep convolutional neural networks for image recognition

A broad range of prior research has demonstrated that attention mechanisms offer great potential in advancing the performance of deep convolutional neural networks (CNNs). However, most existing approaches either ignore modeling attention in both channel and spatial dimensions or introduce higher model complexity and heavier computational burden. To alleviate this dilemma, in this paper, we propose a lightweight and efficient multidimensional collaborative attention, MCA, a novel method for simultaneously inferring attention in channel, height, and width dimensions with almost free additional overhead by using a three-branch architecture. For the essential components of MCA, we not only develop an adaptive combination mechanism for merging dual cross-dimension feature responses in squeeze transformation, enhancing the informativeness and discriminability of feature descriptors but also design a gating mechanism in excitation transformation that adaptively determines the coverage of interaction to capture local feature interactions, overcoming the paradox of performance and computational overhead trade-off. Our MCA is simple yet general and can be easily plugged into various classic CNNs as a plug-and-play module and trained along with the vanilla networks in an end-to-end manner. Extensive experimental results for image recognition on CIFAR and ImageNet-1K datasets demonstrate the superiority of our method over other state-of-the-art (SOTA) counterparts. In addition, we also provide insight into the practical benefits of MCA by visually inspecting the GradCAM++ visualization results. The code is available at https://github.com/ndsclark/MCANet.

One and one make eleven: An interpretable neural network for image recognition

Although non-interpretable (black-box) deep learning models are well known for their accuracy, interpretable deep learning models should be used for high stake decisions, such as: healthcare. In this paper, we present a novel technique of combining the existing state-of-the-art models, and using them as a base model to build an interpretable deep learning model Comb-ProtoPNet. In contrast to usual technique of combining the logits of two (or more) algorithms to form an ensemble algorithm, we combine the algorithms themselves. Our proposed interpretable model applies a prototype layer on top of the convolutional layers of an ensemble base model. We trained and tested our algorithm over the dataset of chest CT-scan images of COVID-19 patients, pneumonia patients and normal people. The use of a certain combination of blocks from two different state-of-the-art models (statistically) significantly improved the accuracy compared to the individual use of the state-of-the-art models as the base models, and this is where the part of the title “One and one make eleven” comes from.

Improvement Airport Security System with Face Recognition Using Neural Network Based on the Arduino Uno Microcontroller

The aim of this study is to implement an algorithm for face recognition, based on the Arduino uno microcontroller. This paper presents it as an airport security system to detect passenger’s face and compare the result with the database of unwanted people. The system combines laser trigger circuit for image capturing and artificial neural network for image recognition. The laser trigger circuit has many benefits such as avoiding camera shakes or taking a picture without a timer. The captured image will be analyzed and processed in MATLAB using an artificial neural network to recognize the passenger’s face from the real captured images after the training phase. Many experiments have been conducted on our face databases with various numbers of iterations. The recognitions’ accuracy and efficiency of the proposed model are 93.33% and 2.67 respectively with 0.696530 seconds as execution time. The result shows the robustness of the developed model in terms of mean squared error, execution time, recognitions’ efficiency and accuracy. The smallest obtained mean squared error is 9.9991e-04 for the training data set and 0.1764 for the testing data set when they are recorded for a modified neural network which makes the developed system more reliable. Finally, the artificial neural network demonstrates the ability to detect the unseen relationship between features belong to the same face.

Elastic Adaptively Parametric Compounded Units for Convolutional Neural Network

The activation function introduces nonlinearity into convolutional neural network, which greatly promotes the development of computer vision tasks. This paper proposes elastic adaptively parametric compounded units to improve the performance of convolutional neural networks for image recognition. The activation function takes the structural advantages of two mainstream functions as the function’s fundamental architecture. The SENet model is embedded in the proposed activation function to adaptively recalibrate the feature mapping weight in each channel, thereby enhancing the fitting capability of the activation function. In addition, the function has an elastic slope in the positive input region by simulating random noise to improve the generalization capability of neural networks. To prevent the generated noise from producing overly large variations during training, a special protection mechanism is adopted. In order to verify the effectiveness of the activation function, this paper uses CIFAR-10 and CIFAR-100 image datasets to conduct comparative experiments of the activation function under the exact same model. Experimental results show that the proposed activation function showed superior performance beyond other functions.

Abstract 139: Semi-automated Classification Of Peripheral Artery Disease Lesion Composition From Multi-Contrast Magnetic Resonance Histology At 9.4 Tesla With A 2 Dimensional Convolutional Neural Network Variational AutoEncoder Algorithm

Introduction: Using artificial intelligence (AI)-assisted methods on MRI histology may help endovascular peripheral artery disease (PAD) treatment planning. Objectives: To evaluate soft vs. hard PAD plaque components and crossability using multi-contrast high-resolution MRI, AI algorithm, and histologic samples. Methods: PAD lesions were harvested from 6 patients (one of each) after amputation and imaged using a 9.4T ultra-high resolution MRI scanner with 3 contrasts: T1-weighted (T1w), T2-weighted (T2w) and Ultrashort Echo Time (UTE). 3D imaging volumes were spatially registered and pseudo-color composites (T1w: red, T2w: green, UTE:blue) were created. Custom 2D variational autoencoder (VAE) classification AI algorithm using convolutional neural networks for image recognition classified and scored cross-sections as lumen occluded with soft tissue (tissue score (TS)=10), with soft and hard tissue (TS=30) and with hard tissue (TS= 50). 2 readers performed visual assessment of crossability (crossable/non-crossable) on preselected MRI slices compared to histologic samples (Fig1a), percent agreement was calculated for inter-rater reliability. Results: MRI multi-contrast registration succeeded for all section (pseudo-colors: smooth muscle cells- pink/red, fatty lesions-green, collagen-dark blue, calcification-black). VAE separated axial sections based on plaque composition (Fig1b), average TS per tissue segments ranged from 10 to 48.1 and corresponded well to the composition of tissues inspected visually.20 slices showed appropriate alignment, for the evaluation of crossability 11 were excluded due to lack of sufficient histological sample, inter-rater reliability was 100% on 9 slices (7 crossable, 2 non-crossable). Conclusion: AI algorithm for image recognition successfully distinguished lesion composition. AI-aided analysis may help rapid evaluation of lesion crossability in planning endovascular interventions.

Flagging cartel participants with deep learning based on convolutional neural networks

Adding to the literature on the data-driven detection of bid-rigging cartels, we propose a novel approach based on deep learning (a subfield of artificial intelligence) that flags cartel participants based on their pairwise bidding interactions with other firms. More concisely, we combine a so-called convolutional neural network for image recognition with graphs that in a pairwise manner plot the normalized bids of some reference firm against the normalized bids of any other firms participating in the same tenders as the reference firm. Based on Japanese and Swiss procurement data, we construct such graphs for both collusive and competitive episodes (i.e when a bid-rigging cartel is or is not active) and we use a subset of graphs to train the neural network such that it learns distinguishing collusive from competitive bidding patterns. With the remaining graphs, we test the neural network’s out-of-sample performance in correctly classifying collusive and competitive bidding interactions. We obtain a very decent average accuracy of around 95% or slightly higher when either applying the method within Japanese, Swiss, or mixed data (in which Swiss and Japanese graphs are pooled). When using data from one country for training to test the trained model’s performance in the other country (i.e. transnationally), predictive performance decreases (likely due to institutional differences in procurement procedures across countries), but often remains satisfactorily high. All in all, the generally quite high accuracy of the convolutional neural network despite being trained in a rather small sample of a few 100 graphs points to a large potential of deep learning approaches for flagging and fighting bid-rigging cartels.

An Image Detection–Memory–Recognition Artificial Visual Unit Based on Dual‐Gate Phototransistors

Optoelectronic synapses integrating sensing–memory–processing functions have great advantages in neuromorphic computing for visual information processing and complex learning, recognition, and memory in an energy‐efficient manner. Herein, a light‐induced bidirectional response is demonstrated in the proposed WSe2/MoS2 junction field‐effect transistor (JFET) with an extra Ge back gate. The WSe2/MoS2 JFET exhibits high responsivity and detectivity owing to effective modulation by the top junction and back dielectric gates. The unique bidirectional photoresponse and interfacial state storage properties of the device result in significant synaptic excitatory effects under visible light stimulation and remarkable synaptic inhibitory effects under infrared illumination. Optical storage with visible light and optical erasure with infrared light can be achieved in the device based on its synaptic behavior. The conductance changes under different types of illumination can be used to mimic the weight update in a neural network for image recognition. A high accuracy exceeding 98% was achieved in handwritten digit recognition. This visible and infrared dual‐band phototransistor, which integrates image perception, memory, and recognition functionalities, provides a promising solution for self‐driving, surveillance, computer vision, and biomedical imaging applications.

A Conic Radon-based Convolutional Neural Network for Image Recognition

This article presents a new approach for image recognition that proposes to combine Conical Radon Transform (CRT) and Convolutional Neural Networks (CNN). In order to evaluate the performance of this approach for pattern recognition task, we have built a Radon descriptor enhancing features extracted by linear, circular and parabolic RT. The main idea consists in exploring the use of Conic Radon transform to define a robust image descriptor. Specifically, the Radon transformation is initially applied on the image. Afterwards, the extracted features are combined with image and then entered as an input into the convolutional layers. Experimental evaluation demonstrates that our descriptor which joins together extraction of features of different shapes and the convolutional neural networks achieves satisfactory results for describing images on public available datasets such as, ETH80, and FLAVIA. Our proposed approach recognizes objects with an accuracy of 96 % when tested on the ETH80 dataset. It also has yielded competitive accuracy than state-of-the-art methods when tested on the FLAVIA dataset with accuracy of 98 %. We also carried out experiments on traffic signs dataset GTSBR. We investigate in this work the use of simple CNN models to focus on the utility of our descriptor. We propose a new lightweight network for traffic signs that does not require a large number of parameters. The objective of this work is to achieve optimal results in terms of accuracy and to reduce network parameters. This approach could be adopted in real time applications. It classified traffic signs with high accuracy of 99%.

Designing a Convolutional Neural Network for Image Recognition: A Comparative Study of Different Architectures and Training Techniques

Designing a Convolutional Neural Network for Image Recognition: A Comparative Study of Different Architectures and Training Techniques

Recent progress in ferroelectric synapses and their applications

Brain-like computing is an important direction for the development of integrated circuits in the post-Moore era, and the development of artificial synaptic devices that can simulate ideal synaptic behavior is the key to building brainlike computing chips with neuron-synapse-neuron connectivity. Ferroelectric thin film materials have unique nonvolatile polarization, and the plasticity of polarization is very similar to that of biological synapses, so ferroelectric synaptic devices have been widely concerned in recent years. In this paper, the research progress of ferroelectric synaptic devices is reviewed from two aspects, simulation of synaptic function in devices and brain-like computing applications. The results show that ferroelectric synaptic devices have two typical structures of two-terminal and three-terminal. In addition to effectively simulating biological synapse functions, ferroelectric synaptic devices also have the advantages of simple structure, low power consumption, high stability, large switching ratio, and fast programming speed. In terms of applications, a series of advances have been made in the study of ferroelectric synapse-based neural networks for image recognition; meanwhile, ferroelectric synapses have also been applied to tactile and visual bionics. Although abundant research progress has been made, ferroelectric synapses are still at the proof-of-principle stage. There are considerable challenges in the synaptic performance regulation mechanism, reliability evaluation criteria, array structure optimization design, high-density integration process, neuromorphic computing architecture design, and novel application scenario expansion, which are also the directions that need to be focused for future ferroelectric synapse research.

Image Target Recognition Based on Improved Convolutional Neural Network

Convolutional neural network (CNN) algorithm is a very important branch of deep learning research, which has been widely applied in many fields and achieved excellent results, especially in computer vision, where convolutional neural network has made breakthroughs in image classification and object detection. Convolutional neural network architecture can realize more efficient network training through the final combination of different modules, and the convolutional neural network training does not need to actively extract image features and can directly carry out end-to-end training and prediction. At first, this paper analyzed some problems of the current image recognition and expounds the progress of convolution neural network in image recognition and then studied the traditional algorithm of target recognition, including traditional recognition algorithm framework of target, the target orientation, feature extraction, classifier classification, etc., and the traditional target recognition algorithm is compared with those of the target recognition algorithm of deep learning. On the basis of the above research, an improved model of CNN is proposed, which focuses on the structural design and network optimization of convolutional neural network and designs a more efficient convolutional neural network. Test experiments verify the effectiveness of the proposed model, which not only achieves lower error rate, but also greatly reduces the number of network parameters and has stronger learning ability.

Structured Binary Neural Networks for Image Recognition

In this paper, we propose to train binarized convolutional neural networks (CNNs) that are of significant importance for deploying deep learning to mobile devices with limited power capacity and computing resources. Previous works on quantizing CNNs often seek to approximate the floating-point information of weights and/or activations using a set of discrete values. Such methods, termed value approximation here, typically are built on the same network architecture of the full-precision counterpart. Instead, we take a new “structured approximation” view for network quantization — it is possible and valuable to exploit flexible architecture transformation when learning low-bit networks, which can achieve even better performance than the original networks in some cases. In particular, we propose a “group decomposition” strategy, termed GroupNet, which divides a network into desired groups. Interestingly, with our GroupNet strategy, each full-precision group can be effectively reconstructed by aggregating a set of homogeneous binary branches. We also propose to learn effective connections among groups to improve the representation capability. To improve the model capacity, we propose to dynamically execute sparse binary branches conditioned on input features while preserving the computational cost. More importantly, the proposed GroupNet shows strong flexibility for a few vision tasks. For instance, we extend the GroupNet for accurate semantic segmentation by embedding the rich context into the binary structure. The proposed GroupNet also shows strong performance on object detection. Experiments on image classification, semantic segmentation, and object detection tasks demonstrate the superior performance of the proposed methods over various quantized networks in the literature. Moreover, the speedup and runtime memory cost evaluation comparing with related quantization strategies is analyzed on GPU platforms, which serves as a strong benchmark for further research.

Development of variational quantum deep neural networks for image recognition

Parametrized quantum circuits are widely used for supervised learning tasks such as image classification in the noisy intermediate scale quantum era. However, normally, it can only handle low-dimensional data. This study presented a variational quantum deep neural network (VQDNN) model for various scale image recognition tasks. Three classifiers were designed to verify the classification performance of the proposed VQDNN model. In the first classifier, to accommodate the limitations of qubits in both simulation hardware and real quantum hardware – we adopted hybrid principal component analysis – VQDNN architecture. Moreover, the amplitude encoding scheme and the rotation angle coding scheme were employed in the subsequent two classifiers to handle large-size images. Finally, we used the classical neural network and VQDNN model to conduct a comparative experiment of the ten-label classification learning task on the same dataset. The quantum numerical experiment was implemented on two benchmark datasets: the MNIST and UCI databases of handwritten digits. The simulation results showed that the proposed VQDNN classified the two datasets with an accuracy of 100% for the two-class classification task, while the UCI dataset has an accuracy of 90.87% for the ten-label classification task. The proposed VQDNN achieved better classification accuracy than the original classical neural network even under a limited number of qubits available in current hardware, indicating the promising application potential of VQDNN in image recognition.

A Machine Learning Strategy for Race-Tracking Detection During Manufacturing of Composites by Liquid Moulding

This work presents a supervised machine learning (ML) model to detect race-tracking disturbances during the liquid moulding manufacturing of structural composites. Race-tracking is generated by unexpected resin channels at mould edges that may induce dry spots and porosity formation. The ML model uses the pressure signals recorded by a sensor network as input, providing a classification of the race-tracking event from a set of possible scenarios, and a subsequent variable regression for their position, size and strength. Such a model is based on the residual network (ResNet), a well-known artificial intelligence architecture that makes use of convolutional neural networks for image recognition. Training of the ML classifier and regressors was carried out with the aid of a synthetically generated simulation data set obtained throughout computational fluid dynamics simulations. The time evolution of the pressure sensors was used as grey-level images, or footprints, as inputs to the ResNet ML. The trained model was able to recognise the presence of race-tracking channels from the pressure data yielding good accuracy in terms of label prediction as well as position, size and strength. The model correlation was carried out with a set of injection experiments performed with a constant thickness closed mould containing induced race-tracking channels. The ability of ML models to provide an approximation to the inverse problem, relating the pressure sensor distortions to the cause of such events, is analysed and discussed.