MinMAE calibration method for convolutional neural network quantization

A multi-resolution fusion approach for human activity recognition from video data in tiny edge devices

Unmanned aerial vehicles (UAV) and related technologies have played an active role in the prevention and control of novel coronaviruses at home and abroad, especially in epidemic prevention, surveillance, and elimination. However, the existing UAVs have a single function, limited processing capacity, and poor interaction. To overcome these shortcomings, we designed an intelligent anti-epidemic patrol detection and warning flight system, which integrates UAV autonomous navigation, deep learning, intelligent voice, and other technologies. Based on the convolution neural network and deep learning technology, the system possesses a crowd density detection method and a face mask detection method, which can detect the position of dense crowds. Intelligent voice alarm technology was used to achieve an intelligent alarm system for abnormal situations, such as crowd-gathering areas and people without masks, and to carry out intelligent dissemination of epidemic prevention policies, which provides a powerful technical means for epidemic prevention and delaying their spread. To verify the superiority and feasibility of the system, high-precision online analysis was carried out for the crowd in the inspection area, and pedestrians’ faces were detected on the ground to identify whether they were wearing a mask. The experimental results show that the mean absolute error (MAE) of the crowd density detection was less than 8.4, and the mean average precision (mAP) of face mask detection was 61.42%. The system can provide convenient and accurate evaluation information for decision-makers and meets the requirements of real-time and accurate detection.

The purposes are to analyze the safety and real-time performance of intelligent transportation under big data and solve urban transportation safety problems. The existing problems of complex traffic data collection and single factor analysis are analyzed. Guangzhou target road section and its historical traffic flow are taken as the data source, the big data analytics (BDA) platform is constructed after preprocessing, and then the deep belief network (DBN) is introduced to build the intelligent transportation model based on BDA. Finally, the model is simulated to analyze its performance. The results reveal that the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) of the proposed algorithm are 4.42%, 6.21%, and 8.03%, respectively compared to Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), Neuro fuzzy c-means model (FCM), Deep Neural Network (DNN), and neural network (NN). The mean absolute error (MAE) and mean absolute percentage error (MAPE) are 4.42%, 6.21% and 8.03% respectively. The prediction accuracy is significantly better than that of other algorithms. The analysis of Accuracy, Precision, Recall and F1 suggests that the safety prediction accuracy of the proposed algorithm can reach up to 88.57%, which is at least 3.19% higher than that of other algorithms. Meantime, the proposed algorithm can effectively inhibit the spread of congestion, and achieve the effect of timely evacuation for traffic congestion. Regarding the data transmission performance, the delay is less than 40ms and the packet loss rate is less than 0.1, which is significantly better than other model algorithms in real-time performance. Hence, the constructed intelligent transportation algorithm based on BDA has high precision and excellent congestion evacuation performance, which can provide an experimental basis for constructing intelligent and safe transportation in the future.

Accurate prediction of agricultural yield is extremely important to ensure food security and cope with the challenges created by climate change and natural disasters. Forecasting agricultural yield is a challenging task due to the complex nature of variables (fertiliser, rainfall, temperature and others) that affect agricultural production. This study employs six supervised machine learning algorithms: Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Multi-Layer Perceptron (MLP), Recurrent Neural Network (RNN), and Convolutional Neural Network (CNN) to build a predictive model using 49 years of historical data (1973-2021) on paddy, wheat, and maize. Model performance was evaluated using Mean Squared Error (MSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Squared Error (rMSE). Results show that DT and RF models are the most precise with MSE 1% to 5%, MAE 8% to 21%, followed by SVM and CNN. Key predictors of crop yield include area cultivated, capital expenditure, banking expansion, rainfall, temperature, and fertilizers, while irrigation and road network were less significant. The study recommends that farmers prioritize commercial farming, agricultural equipment, and timely availability of fertilizer for application. The Government of Nepal (GoN) should redirect subsidies towards agricultural mechanization, ensure timely supply of fertilizer, and expand banking services in agricultural areas.

The compression of deep learning models is of fundamental importance in deploying such models to edge devices. The selection of compression parameters can be automated to meet changes in the hardware platform and application. This article introduces a Multi-Objective Hardware-Aware Quantization (MOHAQ) method, which considers hardware performance and inference error as objectives for mixed-precision quantization. The proposed method feasibly evaluates candidate solutions in a large search space by relying on two steps. First, post-training quantization is applied for fast solution evaluation (inference-only search). Second, we propose the ”beacon-based search” to retrain selected solutions only and use them as beacons to estimate the effect of retraining on other solutions. We use speech recognition models on TIMIT dataset. Experimental evaluations show that Simple Recurrent Unit (SRU)-based models can be compressed up to 8x by post-training quantization without any significant error increase. On SiLago, we found solutions that achieve 97% and 86% of the maximum possible speedup and energy saving, with a minor increase in error on an SRU-based model. On Bitfusion, the beacon-based search reduced the error gain of the inference-only search on SRU-based models and Light Gated Recurrent Unit (LiGRU)-based model by up to 4.9 and 3.9 percentage points, respectively.

Considerable attention has recently been focused on classification and regression-based convolutional neural network (CNN) and long short-term memory (LSTM) due to their excellent performance in capturing complex spatial and temporal information characteristics for structural damage identification. However, few studies have considered structural damage identification as a classification and regression problem. In addition, bridges in practical engineering are vulnerable to various environmental and vehicle loading conditions. Hence, this study proposed a new two-stage CNN–LSTM configuration for bridge damage identification using vibration data considering the influence of temperatures. First, a classification-based CNN–LSTM is designed to perform multiclass damage detection tasks, and then a regression-based CNN–LSTM is developed for damage localization and severity prediction tasks. The performance of the proposed damage identification method was evaluated through a simulation dataset of a concrete highway bridge model and a field experiment dataset of Z24-bridge (Switzerland). In addition, a set of statistical evaluation metrics such as sparse categorical cross-entropy loss, accuracy, confusion matrix, mean squared loss, mean absolute error, mean absolute percentage error, and coefficient of determination were used to compare the damage identification performance of the proposed CNN–LSTM configuration with a regular CNN model and conventional machine learning (ML) algorithms. Prediction results indicate that the proposed CNN–LSTM model outperforms the regular CNN model and conventional ML algorithms for bridge damage identification.

To develop and evaluate an artificial intelligence (AI) system that can automatically calculate the glomerular filtration rate (GFR) from dynamic renal imaging without manually delineating the regions of interest (ROIs) of kidneys and the corresponding background. This study was a single-center retrospective analysis of the data of 14,634 patients who underwent 99mTc-DTPA dynamic renal imaging. Two systems based on convolutional neural networks (CNN) were developed and evaluated: sGFRa predicts the radioactive counts of ROIs and calculates GFR using the Gates equation and sGFRb directly predicts GFR from dynamic renal imaging without using other information. The root-mean-square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and R2 were used to evaluate the performance of our approach. sGFRa achieved an RMSE of 5.05, MAE of 4.03, MAPE of 6.07%, and R2 of 0.93 for total GFR while sGFRb achieved an RMSE of 7.61, MAE of 5.92, MAPE of 8.92%, and R2 of 0.85 for total GFR. The accuracy of sGFRa and sGFRb in determining the stage of chronic kidney disease was 87.41% and 82.44%, respectively. The findings of sGFRa show that automatic GFR calculation based on CNN and using dynamic renal imaging is feasible and efficient and, additionally, can aid clinical diagnosis. Furthermore, the promising results of sGFRb demonstrate that CNN can predict GFR from dynamic renal imaging without additional information. • Our CNN-based AI systems can automatically calculate GFR from dynamic renal imaging without manually delineating the ROIs of kidneys and the corresponding background. • sGFRa accurately predicted the radioactive counts of ROIs and calculated GFR using the Gates method. • sGFRb-predicted GFR directly without any parameters related to the Gates equation.

A CNN encoder decoder LSTM model for sustainable wind power predictive analytics

Forecasting stock prices has always been a complex and challenging task due to its high volatility, noise in data, and abrupt price fluctuations caused by different factors. Traditional approaches, such as quantitative, fundamental, and technical analysis, often struggle to accurately predict stock price movements due to their difficulty in processing non-linear data. The evolution of machine learning, particularly neural network models, opened new opportunities for improving prediction accuracy by handling non-linear and multidimensional data with high effectiveness. This study focuses on evaluating and comparing three neural network architectures, in which the last is a combination of the first two - Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and a hybrid CNN-LSTM model, for forecasting stock prices. Using ten years of historical stock data, taken for Apple Inc. (AAPL) stocks as an example, ranging from 2015 to 2025, the models were trained and tested on key performance metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and R². Results show that while both CNN and LSTM models show good performance individually, the hybrid CNN-LSTM model can significantly outperform them, achieving the highest accuracy with the lowest error rates. By combining the strengths of CNN for feature extraction and LSTM for temporal sequence processing, the hybrid model effectively captures complex patterns in stock price movements. By iterating over the types of layers to use, we came to the conclusion, that by combining both models into one hybrid model, some layers, like one CNN layer, and one LSTM layer should be removed. This research demonstrates the potential of hybrid neural network models in financial time-series forecasting and comprehensively analyzes their advantages over non-hybrid models.

In this work, we present an approach based on multilevel local as well as global Convolutional Neural Network (CNN) feature matching to retrieve near duplicate images. CNN features are suitable for visual matching. The CNN features of entire image may not give accuracy in retrieval due to various image editing/capturing operations. Our retrieval task focuses on matching image pairs based on local and global levels. In local matching, an image is segmented into fixed size blocks followed by extracting patches by considering neighboring regions at different levels. Matching local image patches at different levels provides robustness to our retrieval model. In local patch extraction, we select blocks containing SURF feature points instead of selecting all blocks. CNN features are extracted and stored for each image patch and then followed by extraction of global CNN features. Finally, similarity between image pairs is computed by considering all extracted CNN features. Our similarity function is based on correlation and number of blocks found in matching. We implemented our proposed approach on benchmarking Holiday dataset. Retrieval results show remarkable improvement in mean average precision (mAP) on the dataset.

Estimating the State of Health (SOH) of lithium batteries is vital for the safe management of new energy systems. This study leverages voltage, current, and temperature data from the NASA battery dataset to extract health features. The relationship between these features and battery capacity is evaluated using mutual information analysis. An integrated Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), and Squeeze-and-Excitation (SE) model (CNN-GRU-SE) for estimating battery SOH is proposed. The CNN module identifies local features from the input data, the SE module emphasizes key features, and the GRU module captures temporal dependencies, effectively tracking the battery’s health trend over time. The estimation results indicate that the CNN-GRU-SE model reduces the Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE) by approximately 3% to 10% compared to the CNN, GRU, and CNN-GRU models. These results confirm the superior estimation capability of the integrated CNN-GRU-SE model. Furthermore, the study underscores the effective integration of the strengths of CNN, SE modules, and GRU, demonstrating its potential application in battery health management.

Accurate electrical load forecasting is of great significance to help power companies in better scheduling and efficient management. Since high levels of uncertainties exist in the load time series, it is a challenging task to make accurate short-term load forecast (STLF). In recent years, deep learning approaches provide better performance to predict electrical load in real world cases. The convolutional neural network (CNN) can extract the local trend and capture the same pattern, and the long short-term memory (LSTM) is proposed to learn the relationship in time steps. In this paper, a new deep neural network framework that integrates the hidden feature of the CNN model and the LSTM model is proposed to improve the forecasting accuracy. The proposed model was tested in a real-world case, and detailed experiments were conducted to validate its practicality and stability. The forecasting performance of the proposed model was compared with the LSTM model and the CNN model. The Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) were used as the evaluation indexes. The experimental results demonstrate that the proposed model can achieve better and stable performance in STLF.

Magnetic resonance imaging (MRI)-guided radiation therapy (RT) treatment planning is limited by the fact that the electron density distribution required for dose calculation is not readily provided by MR imaging. We compare a selection of novel synthetic CT generation algorithms recently reported in the literature, including segmentation-based, atlas-based and machine learning techniques, using the same cohort of patients and quantitative evaluation metrics. Six MRI-guided synthetic CT generation algorithms were evaluated: one segmentation technique into a single tissue class (water-only), four atlas-based techniques, namely, median value of atlas images (ALMedian), atlas-based local weighted voting (ALWV), bone enhanced atlas-based local weighted voting (ALWV-Bone), iterative atlas-based local weighted voting (ALWV-Iter), and a machine learning technique using deep convolution neural network (DCNN). Organ auto-contouring from MR images was evaluated for bladder, rectum, bones, and body boundary. Overall, DCNN exhibited higher segmentation accuracy resulting in Dice indices (DSC) of 0.93±0.17, 0.90±0.04, and 0.93±0.02 for bladder, rectum, and bones, respectively. On the other hand, ALMedian showed the lowest accuracy with DSC of 0.82±0.20, 0.81±0.08, and 0.88±0.04, respectively. DCNN reached the best performance in terms of accurate derivation of synthetic CT values within each organ, with a mean absolute error within the body contour of 32.7±7.9HU, followed by the advanced atlas-based methods (ALWV: 40.5±8.2HU, ALWV-Iter: 42.4±8.1HU, ALWV-Bone: 44.0±8.9HU). ALMedian led to the highest error (52.1±11.1HU). Considering the dosimetric evaluation results, ALWV-Iter, ALWV, DCNN and ALWV-Bone led to similar mean dose estimation within each organ at risk and target volume with less than 1% dose discrepancy. However, the two-dimensional gamma analysis demonstrated higher pass rates for ALWV-Bone, DCNN, ALMedian and ALWV-Iter at 1%/1mm criterion with 94.99±5.15%, 94.59±5.65%, 93.68±5.53% and 93.10±5.99% success, respectively, while ALWV and water-only resulted in 86.91±13.50% and 80.77±12.10%, respectively. Overall, machine learning and advanced atlas-based methods exhibited promising performance by achieving reliable organ segmentation and synthetic CT generation. DCNN appears to have slightly better performance by achieving accurate automated organ segmentation and relatively small dosimetric errors (followed closely by advanced atlas-based methods, which in some cases achieved similar performance). However, the DCNN approach showed higher vulnerability to anatomical variation, where a greater number of outliers was observed with this method. Considering the dosimetric results obtained from the evaluated methods, the challenge of electron density estimation from MR images can be resolved with a clinically tolerable error.

Sales forecasting is essential for businesses to predict future demand and inform strategic and operational planning, especially in the building materials retail industry. Accurate sales prediction supports inventory management, cost control, and supply chain efficiency. This study compares the performance of 3 deep learning models, Convolutional Neural Network (CNN), Multilayer Perceptron (MLP), and Long Short-Term Memory (LSTM), in forecasting daily iron sales at PT Surya Aneka Bangunan from 2016 to 2020. The models were trained on 80% of the historical data and tested on 20%. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Coefficient of Determination R². The results show that the CNN model achieved the best performance with an MAE of 0.293, RMSE of 0.357, MAPE of 0.081, and R² of 0.9989, indicating high accuracy and stability. The MLP model produced higher errors, while the LSTM model had the lowest MAPE but greater error variability. These findings suggest that the CNN model is the most reliable for capturing temporal patterns in iron sales data. The study contributes to the development of adaptive sales forecasting systems and opens opportunities for applying similar methods in other retail sectors to support data driven decision making.

The overheating of heat transfer wall is a primary contributor to boiler tube failures in coal-fired boilers. A fusion model that integrates sparse self-attention (SSA), convolutional neural networks (CNN), and bidirectional long short-term memory networks (BiLSTM) is proposed to predict the wall temperature of the final reheater, based on their dynamic characteristics. Firstly, the original candidate variables are subjected to filtering and dimensionality reduction through the KPCA algorithm, resulting in the selection of the top 26 principal component variables as the final input for the model. Secondly, leveraging the strengths of convolutional neural networks (CNN) in capturing local correlations and bidirectional long short-term memory networks (BiLSTM) in modeling long-term sequential dependencies, a CNN-BiLSTM architecture is employed to effectively capture both short-term and long-term dependencies within time series data. Additionally, a sparse self-attention mechanism (SSA) is incorporated to enhance CNN-BiLSTM’s performance by assigning varying weights to different components based on their significance. Simulations are performed utilizing historical data from an operational 1000MW ultra-supercritical boiler. The results indicate that the root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) for the CNN-BiLSTM-SSA model are 4.92°C, 3.81°C, and 0.624%, respectively, demonstrating superior performance compared to CNN, LSTM, BiLSTM, CNN-LSTM, and CNN-BiLSTM models. The presented deep learning fusion model effectively predicts high-temperature reheater wall temperatures and holds significant theoretical and engineering value for addressing overheating issues in advance.