Angel-Eye: A Complete Design Flow for Mapping CNN Onto Embedded FPGA

Among the many deep learning methods, the convolutional neural network (CNN) model has an excellent performance in image recognition. Research on identifying and classifying image datasets using CNN is ongoing. Animal species recognition and classification with CNN is expected to be helpful for various applications. However, sophisticated feature recognition is essential to classify quasi-species with similar features, such as the quasi-species of parrots that have a high color similarity. The purpose of this study is to develop a vision-based mobile application to classify endangered parrot species using an advanced CNN model based on transfer learning (some parrots have quite similar colors and shapes). We acquired the images in two ways: collecting them directly from the Seoul Grand Park Zoo and crawling them using the Google search. Subsequently, we have built advanced CNN models with transfer learning and trained them using the data. Next, we converted one of the fully trained models into a file for execution on mobile devices and created the Android package files. The accuracy was measured for each of the eight CNN models. The overall accuracy for the camera of the mobile device was 94.125%. For certain species, the accuracy of recognition was 100%, with the required time of only 455 ms. Our approach helps to recognize the species in real time using the camera of the mobile device. Applications will be helpful for the prevention of smuggling of endangered species in the customs clearance area.

Edge AI systems are increasingly being adopted in a wide range of application domains. These systems typically deploy Convolutional Neural Network (CNN) models on edge devices to perform inference, while relying on the cloud for model training. This is due to the high computational and memory demands of conventional model training, which exceeds the capabilities of resource-constrained edge devices running on tight power budgets. The dependency on the cloud for training is not suitable in many applications, where new objects or environmental conditions different from the ones present during training, are frequently encountered. In such applications, continual learning of new knowledge on the edge device becomes a necessity to avoid performance bottlenecks due to round-trip communication delays, network connectivity, and the available bandwidth. In this thesis, we propose Field-Programmable Gate Array (FPGA) based accelerator architecture and optimization strategies for a new paradigm of machine learning algorithms that is capable of continual learning. The proposed methods will enable edge FPGA systems to perform on-device deep continual learning for object classification. Specifically, the proposed methods aim to achieve real-time learning on-device, while providing for a high degree of scalability to learn a large number of classes. We first propose a FPGA accelerator for a Self-Organizing Neural Network (SONN), that can perform class-incremental continual learning in a streaming manner when combined with a CNN. The SONN model performs unsupervised learning from embedding features extracted from the CNN model by dynamically growing neurons and connections. We introduce design optimization strategies and runtime scheduling techniques to optimize resource usage, latency, and energy consumption. Experimental results based on Core50 dataset for continuous object recognition from video sequences, demonstrated that the proposed FPGA architecture outperforms CPU and GPU based counterparts in terms of latency and power. However, the SONN model grows proportionally to the classes learnt, which limits its scalability to learn a large number of classes efficiently. Next, we propose a FPGA accelerator for a Streaming Linear Discriminant Analysis (SLDA) model to overcome the scalability limitations of SONN. Similar to the SONN, SLDA performs continual learning from embedding features extracted from a CNN in a streaming manner. SLDA is highly scalable for learning a large number of classes as the network does not grow dynamically to accommodate new knowledge. We propose several design and runtime optimizations to minimize resource usage, latency, and energy consumption. Additionally, we introduce a new variant of SLDA and discuss the accuracy-efficiency trade-offs using popular datasets for continual learning, CoRE50 and CUB200. The results demonstrate that the proposed SLDA accelerator outperforms CPU and GPU counterparts, in terms of latency and energy efficiency. Finally, we demonstrate a full on-chip deep continual learning pipeline on FPGA, by integrating the proposed SLDA accelerator with Xilinx DPU, a programmable CNN accelerator IP. The design is implemented on a Xilinx Zynq Ultrascale+ MPSoC. In order to overcome the large performance bottleneck due to the communication overhead between the ARM processing system (PS) and programmable logic (PL), we implemented a Linux device driver that facilitates efficient memory mapping between PS-PL. The experimental results based on CoRE50 dataset show that the proposed pipeline is capable of performing continual learning at nearly the same latency as the inference pipeline, with only a marginal increase in energy consumption. Our results clearly demonstrate the viability of deploying real-time deep continual learning on edge AI systems that are equipped with FPGA accelerators.

Understanding the learning mechanism of convolutional neural networks in spectral analysis

Accurately identifying epileptogenic zone (EZ) using high-frequency oscillations (HFOs) is a challenge that must be mastered to transfer HFOs into clinical use. We analyzed the ability of a convolutional neural network (CNN) model to distinguish EZ and non-EZ HFOs. Nineteen medically intractable epilepsy patients with good surgical outcomes 2 years after surgery were studied. Five-minute interictal intracranial electroencephalogram epochs of slow-wave sleep were selected randomly. Then 5 s segments of ripples (80–200 Hz) and fast ripples (FRs, 200–500 Hz) were detected automatically. The EZs and non-EZs were identified using the surgery resection range. We innovatively converted all epochs into four types of images using two scales: original waveforms, filtered waveforms, wavelet spectrum images, and smoothed pseudo Wigner–Ville distribution (SPWVD) spectrum images. Two scales were fixed and fitted scales. We then used a CNN model to classify the HFOs into EZ and non-EZ categories. As a result, 7,000 epochs of ripples and 2,000 epochs of FRs were randomly selected from the EZ and non-EZ data for analysis. Our CNN model can distinguish EZ and non-EZ HFOs successfully. Except for original ripple waveforms, the results from CNN models that are trained using fixed-scale images are significantly better than those from models trained using fitted-scale images (p < 0.05). Of the four fixed-scale transformations, the CNN based on the adjusted SPWVD (ASPWVD) produced the best accuracies (80.89 ± 1.43% and 77.85 ± 1.61% for ripples and FRs, respectively, p < 0.05). The CNN using ASPWVD transformation images is an effective deep learning method that can be used to classify EZ and non-EZ HFOs.

The convolutional neural network (CNN) has been widely applied in computer vision applications. Due to the intensive computation, the general central processing unit (CPU) processors are not efficient to meet the real-time requirement. Various hardware accelerators based on the application specific integrated circuit (ASIC) and field programmable gate array (FPGA) have been designed to accelerate CNN models during the inference phase. The data flow architecture has been extensively adopted because of the high parallelism. However, given the continual development in the computer vision field, CNN models have become increasingly diverse. Thus, CNN accelerators based on data flow architectures face an emerging challenge to maintain high throughput while coping with various CNN models. In this paper, we design a software-defined architecture to solve this. The goal of this study is to make the hardware change as the application changes to achieve high flexibility and high performance. In our proposed architecture, all the parts can be software-defined to cope with different CNN models. A flexible software-defined processing element (PE) array is designed to compute different weight filter sizes. In addition, a software-defined data reuse technique based on two ideal reuse cases is proposed to ensure that all the parameters need to be loaded only once during the computing phase. To support this reuse technique, we also propose the software-defined on-chip buffer so that the weight and image buffers share one dynamic buffer. By using the sparsity property of the input feature map, the full-connected (FC) layer is accelerated. About 88% of the FC weight parameters can be skipped when loading the VGG-16 model. Finally, we implemented this software-defined accelerator on the FPGA. Compared to the other FPGA based accelerators, our proposed accelerator can preserve high performance while maintaining flexibility.

Distinguishing pericarpium citri reticulatae of different origins using terahertz time-domain spectroscopy combined with convolutional neural networks

Convolutional Neural Networks (CNNs) have demonstrated impressive performance in complex machine learning tasks such as classification and regression problems. A reliable neural network structure plays a decisive role in CNN studies. Through comparing and analyzing the structure of neural networks, a model structure for better visualization performance has been discovered, and such a method supports the development of deep learning research. These studies are of particular importance in end-to-end systems for autonomous driving to imitate human driving, where the interpretability of the system is limited. Because of the uncertainty of the ground truth, for the determination of human steering in an image, it is difficult to accurately compare the visualization performance of different CNN models or different visualization methods. For practical applications, however, an objective and quantitative measure for assessing visualization performance is necessary. Therefore, a method to evaluate the visualization performance of CNN models using a driver model instead of human drivers is proposed, to generate a data set which can be used to determine the decisional point (ground truth) in the input image. Then, an exclusive method is also put forth, to quantitatively calculate the relationship between the decisional point (ground truth) and the visualization results produced by CNN models. In this paper, five CNN models as an autonomous steering controller are designed based on PilotNet, and the visualization abilities of each CNN models is compared by three evaluation indicators. By comparing the visualization performance of five different CNN models, it is shown that the proposed method can successfully assess the visualization level of the CNN model.

Identification of crops is an important topic in the agricultural domain. Hyperspectral remote sensing data are very useful for crop feature extraction and classification. Remote sensing data is an unstructured data and Convolutional Neural Network (CNN) can work on unstructured data efficiently. This paper presents an evaluation of CNN for crop classification using the Indian Pines standard dataset obtained from the AVIRIS sensor and the study area dataset obtained from the EO-1hyperion sensor. Optimized CNN has been tuned by training the model on different parameters. It has been compared with two classification algorithms: Deep Neural Network (DNN) and Convolutional Autoencoder. According to the test results, the proposed optimized CNN model provided better results as compared to the other two methods. CNN has given 97 ± 0.58% overall accuracy for the Indian Pines standard dataset and 78 ± 2.43% for our study area dataset.

The Part 1 prequel to this review evaluated the evolution of modelling techniques used in evaluation of fruit quality over the past three decades and noted a progression towards the use of artificial neural networks (ANNs) and convolutional neural networks (CNNs). In this review, Part 2, the use of CNNs for NIR fruit quality evaluation is explored, given the success of CNNs in various other fields, such as image, video, speech, and audio processing, and the availability of large (open source) datasets of fruit spectra and reference quality attribute, which is required for the training of CNN models. The review provides an overview of deep learning and the CNN architectures and techniques used in NIR spectroscopy for regression modelling, with advantages and disadvantages identified. Studies using CNN for NIR based fruit quality evaluation are then critically examined. Eight publications have presented on models using the same open-source mango dry matter calibration and test set, enabling inter-method comparisons. CNN models have been demonstrated to be accurate, precise and robust. Techniques of transfer learning for CNN models offer an alternative solution to model updating and calibration transfer methods applied in traditional chemometrics. The review has highlighted crucial areas that require resolution and exploration in this application through future research, including, (i) data requirements for training a CNN (ii) optimal spectral pre-processing for CNN (iii) CNN architecture and hyper-parameter selection and tuning for fruit quality evaluation (iv) CNN model interpretability and explainability. Future studies must conduct clearer comparison to partial least squares (PLS) regression and shallow ANNs to better assess the prospective benefit of using CNN, a more complex model. The potential for visualisation of spectra relevance to the CNN model using techniques such as GradCam, currently employed in visualising 2D-CNN models, remains to be explored.

Empirical-statistical Downscaling (SD) techniques are valuable tools able to generate high-resolution climate information needed to carry out impact studies. In this regard, Convolutional Neural Networks (CNNs) are promising SD techniques capable of handling large amounts of data and extracting relevant predictor information for each particular site. These characteristics of the CNN represent a major advantage over traditional SD methods, which typically rely on human-guided predictor selection. Notwithstanding, an adequate tuning of the CNN is key for optimising their potential.In southern South America (SSA), CNNs has proven to be skilful in representing daily extreme temperatures and extrapolating into future scenarios. Although the selection of the activation function introduces a source of uncertainty in the future projections.&amp;#160;In this context, this study aims to explore the use of CNNs as a statistical downscaling tool to simulate the wet bulb temperature (Tw) over SSA, a multivariate heat-stress index estimated from temperature and humidity. Tw has been widely used as a heat-stress proxy in different parts of the world, however, its characterisation and modelling in SSA remain as a pending task. To this end, four different CNN architectures regarding the activation function (ReLU or linear), domain size and configuration of the CNN layers were tested. All CNN models were trained during summer days using a cross-validation (CV) scheme in the period 1991-2020 and then evaluated in four unseen summers between 2021 and 2024. For comparison purposes, CNN models were similarly trained and validated to simulate maximum temperature (Tx).&amp;#160;Overall, CNN models well represented all the features evaluated, including the heat-waves that took place in the summers evaluated independently. In particular, CNN models presents a better performance in simulating Tw than Tx with smaller errors in terms of mean and extremes aspects. Regarding the domain size, for both temperatures, the configuration with the smaller domain yields the best results. Also in this latter case, the reduction of the number of filter size in the last layer slightly improves the representation of Tx. When considering the large domain, the differences between the CNNs based on different activation functions increase, and CNN models with linear configuration outperform the ones with ReLu.&amp;#160;The findings of this work reinforces the potential of CNNs for climate downscaling in SSA, especially for its use to simulate multivariate impact indices.

BACKGROUND AND AIM: Outdoor ultrafine particles (UFP; 100 nm) and black carbon (BC) vary greatly within cities and may have adverse impacts on human health such as cardiovascular mortality and brain tumour incidence. Traditional regression methods for estimating spatial distributions of ambient air pollution depend on extensive, curated geospatial information system databases. This has resulted in a disparity wherein little is known about local air pollution in the data-sparse settings of low- and middle- income countries. We used a hybrid approach to develop new models to estimate within-city spatial variations in outdoor UFP and BC concentrations across Bucaramanga, Colombia. METHODS: We conducted a large-scale mobile monitoring campaign over twenty days in 2019. Land use regression (LUR) models were developed using land use parameters from curated and open-source databases. Convolutional Neural Network (CNN) models were developed using Google Maps satellite view and street view images. Combined models were developed by combining predictions from the LUR and CNN models. After training in the training and validation sets, predictions were generated in the test set and compared to measured values. RESULTS:The combined UFP model (R2=0.54) outperformed the CNN (R2=0.47) and land use regression (LUR) models (R2=0.47) on their own. Similarly, the combined BC model also outperformed the CNN and LUR BC models (R2 = 0.51 vs 0.43 and 0.45 respectively). Spatial variations in model performance were more stable for the CNN and combined models compared to the LUR models suggesting that the combined approach may be less likely to contribute to differential exposure measurement error in epidemiological studies. CONCLUSIONS:Estimates from these models can then be applied to population-based cohorts in order to evaluate population health risks. Additionally, our findings demonstrated that satellite and street-level images can be combined with a traditional LUR modelling approach to improve predictions of within-city spatial variations in outdoor UFP and BC concentrations. KEYWORDS: Ultrafine particles, Black Carbon, Deep learning, Images, Land use regression

This study compares the 5-day tropical cyclone (TC) track forecasts from the Global Ensemble Forecast System (GEFS) with a machine learning approach applied to this ensemble. A convolutional neural network (CNN) model was developed using northern Atlantic TC training data from 2008 to 2022, while an independent set of storms was used to evaluate the CNN model. The CNN for tropical cyclone track forecasting was trained using fivefold cross validation on a shuffled dataset split into 90% training/validation and 10% evaluation. PyTorch was used to develop a CNN model with a custom loss function, and performance was assessed using the haversine function and error metrics, comparing the CNN’s TC track forecasts to GEFS mean track forecasts. The CNN has a 58%–86% better track prediction against GEFS mean for forecast hours 0–96, which decreases to 35%–53% for hours 108–120. Track differences between those cases that the CNN improved the forecast versus did not improve are also explored, which shows that most of the CNN improvement is from a decrease in the along-track error (ATE). This is consistent with past studies of this ensemble, which showed that the largest bias exists in the along-track direction. Significance Statement This research focuses on enhancing the accuracy of tropical cyclone track forecasts by leveraging a convolutional neural network (CNN). Traditional forecasting models have relatively large errors in predicting the track of tropical cyclones, especially at forecast lead times of 3–5 days. These errors can result in inadequate warnings and misallocation of resources during storm preparations, potentially endangering lives and property. Our CNN model addresses these issues by correcting systematic biases, resulting in more accurate forecasts.

Handwritten digit recognition, a task in computer vision, is critical for applications such as postal automation, banking, and digitization of forms. Traditional approaches have leveraged statistical models, but the rise of deep learning, particularly Convolutional Neural Networks (CNN) and Long-Short Term Memory (LSTM), has revolutionized the field. However, a comprehensive performance comparison of CNN and LSTM architectures in the context of handwritten digit classification remains underexplored. This study aimed to address this gap by evaluating and comparing CNN and LSTM models for the classification of handwritten digits. Two machine learning models – Convolutional Neural Network (CNN) and Long-Short Term Memory (LSTM) were trained with the preprocessed handwritten digit dataset. The CNN model was designed with multiple convolutional and pooling layers, along with dropout for regularization. The LSTM model was designed with LSTM layers to capture sequential patterns in the data, followed by a dense layer for classification. The models were implemented in python, evaluated and compared based on accuracy, precision, recall and F1-score. The evaluation and comparison results indicate that CNN achieved 99.31% accuracy, 99.0% precision, 99.0% recall, and a 99.0% F1-score, while LSTM achieved 98.90% accuracy, 99.0% precision, 99.0% recall, and a 99.0% F1-score. The results demonstrated that CNN outperformed LSTM in terms of accuracy and misclassification errors, making it the optimal choice for image-based handwritten digit recognition. This finding underscores the efficiency of CNN in addressing challenges related to digit recognition, contributing to the advancement of automated digit classification systems and improving the accuracy of image-based classification tasks.

Defect inspection is a crucial part of industrial manufacturing. However, it relies heavily on human effort on manual visual inspection. Various machine vision techniques have been introduced to replace human labour and to improve inspection quality and efficiency. The limitation of these techniques is that the algorithms need to be engineered again with each different use case. In this work, a Convolutional Neural Network (CNN) is used to classify the defects of the Chemical Mechanical Planarization (CMP) ring. The trained CNN model achieved an accuracy of 91% and the time taken for each inference process is around 1800 msec. To achieve computational efficiency, the CNN model is performed on the embedded device. The previous implementation of embedded CNN deploys OpenCL-based high-level synthesis accelerator on a high-end Field Programmable Gate Array (FPGA) board. In this work, the model inference is accelerated by PipeCNN FPGA implementation on Cyclone-VSE DE1-SoC, a low-end embedded FPGA board. Several configurations of hardware parameters are tested to search for the best setup of the FPGA resources. The hardware implementation has improved approximately seven times faster, as the inference time for each classification has improved from 1800 msec to 250 msec. However, the model implemented using the hardware is observed to produce lower inference accuracy as the accuracy drops from 91% to 81%.

Excessive withdrawal of groundwater for agricultural irrigation can cause seawater intrusion into coastal aquifers. Such a case will in turn results in deterioration of irrigation water quality. Determination of irrigation water quality with traditional methods is a time-consuming and costly process. However, machine learning algorithms can be useful tools for modeling and estimating groundwater quality used for irrigation water purposes. In this study, TDS, PS, SAR, and Cl parameters of groundwater were estimated with models based on EC and pH variables. For this purpose, prediction performances of two different deep learning methods (convolutional neural network (CNN) and deep neural network (DNN)) and two different classical machine learning (Random Forest (RF) and extreme gradient boosting (XGBoost)) methods were compared. In addition, predictive uncertainty of the models was determined by quantile regression (QR) analysis. Performance criteria and results of uncertainty analysis revealed that CNN (in testing phase, NSE = 0.95 for TDS, NSE = 0.96 for PS, NSE = 0.67 for SAR and NSE = 0.93 for CI) and DNN (in testing phase, NSE = 0.91 for TDS, NSE = 0.91 for PS, NSE = 0.57 for SAR and NSE = 0.94 for Cl) models had quite a close performance in estimation of TDS, PS, SAR, and Cl parameters and higher than the other two classical machine learning methods. As a result, the CNN model can be considered the best performing model in estimating all quality parameters due to the highest NSE and lowest RMSE values. In addition, the Taylor diagram showed that the values estimated using the CNN model had the highest correlation with the measured data. It was determined that the model with the lowest uncertainty based on the PICP statistics was DNN, followed by the CNN model. However, the CNN model has predicted outliers more accurately. Present findings proved that deep learning models could offer efficient tools for predicting irrigation water quality parameters.