Automatic Receipt Recognition System Based on Artificial Intelligence Technology

In this study, an automatic receipt recognition system (ARRS) is developed. First, a receipt is scanned for conversion into a high-resolution image. Receipt characters are automatically placed into two categories according to the receipt characteristics: printed and handwritten characters. Images of receipts with these characters are preprocessed separately. For handwritten characters, template matching and the fixed features of the receipts are used for text positioning, and projection is applied for character segmentation. Finally, a convolutional neural network is used for character recognition. For printed characters, a modified You Only Look Once (version 4) model (YOLOv4-s) executes precise text positioning and character recognition. The proposed YOLOv4-s model reduces downsampling, thereby enhancing small-object recognition. Finally, the system produces recognition results in a tax declaration format, which can upload to a tax declaration system. Experimental results revealed that the recognition accuracy of the proposed system was 80.93% for handwritten characters. Moreover, the YOLOv4-s model had a 99.39% accuracy rate for printed characters; only 33 characters were misjudged. The recognition accuracy of the YOLOv4-s model was higher than that of the traditional YOLOv4 model by 20.57%. Therefore, the proposed ARRS can considerably improve the efficiency of tax declaration, reduce labor costs, and simplify operating procedures.

YOLOv4-Lite–Based Urban Plantation Tree Detection and Positioning With High-Resolution Remote Sensing Imagery

Automatic tree identification and position using high-resolution remote sensing images are critical for ecological garden planning, management, and large-scale environmental quality detection. However, existing single-tree detection methods have a high rate of misdetection in forests not only due to the similarity of background and crown colors but also because light and shadow caused abnormal crown shapes, resulting in a high rate of misdetections and missed detection. This article uses urban plantations as the primary research sample. In conjunction with the most recent deep learning method for object detection, a single-tree detection method based on the lite fourth edition of you only look once (YOLOv4-Lite) was proposed. YOLOv4’s object detection framework has been simplified, and the MobileNetv3 convolutional neural network is used as the primary feature extractor to reduce the number of parameters. Data enhancement is performed for categories with fewer single-tree samples, and the loss function is optimized using focal loss. The YOLOv4-Lite method is used to detect single trees on campus, in an orchard, and an economic plantation. Not only is the YOLOv4-Lite method compared to traditional methods such as the local maximum value method and the watershed method, where it outperforms them by nearly 46.1%, but also to novel methods such as the Chan-Vese model and the template matching method, where it outperforms them by nearly 26.4%. The experimental results for single-tree detection demonstrate that the YOLOv4-Lite method improves accuracy and robustness by nearly 36.2%. Our work establishes a reference for the application of YOLOv4-Lite in additional agricultural and plantation products.

A Precision Functional Atlas of Network Probabilities and Individual-Specific Network Topography

The brain is organized into a broad set of functional neural networks. These networks and their various characteristics have been described and scrutinized through in vivo resting state functional magnetic resonance imaging (rs-fMRI). While the basic properties of networks are generally similar between healthy individuals, there is vast variability in the precise topography across the population. These individual differences are often lost in population studies due to population averaging which assumes topographical uniformity. We leveraged precision brain mapping methods to establish a new open-source, method-flexible set of probabilistic functional network atlases: the Masonic Institute for the Developing Brain (MIDB) Precision Atlas. Using participants from the Adolescent Brain Cognitive Development (ABCD) study, single subject precision maps were generated with two supervised network-matching procedures (template matching and non-negative matrix factorization), as well as an unsupervised community detection algorithm (Infomap). We demonstrate that probabilistic network maps generated for two demographically-matched groups of n~3000 each were nearly identical, both between groups (Pearson r >0.999) and between methods (r=0.96), revealing both regions of high invariance and high variability. Compared to using parcellations based on groups averages, the MIDB Precision Atlases allowed us to derive a set of brain regions that are largely invariant in network topography and provide more reproducible statistical maps of executive function brain-wide associations. We explore an example use case for probabilistic maps, highlighting their potential for use in targeted neuromodulation. The MIDB Precision Atlas is expandable to alternative datasets and methods and is provided open-source with an online web interface to encourage the scientific community to experiment with probabilistic atlases and individual-specific topographies to more precisely relate network phenomenon to functional organization of the human brain.

Template Matching and Matrix Profile for Signal Quality Assessment of Carotid and Femoral Laser Doppler Vibrometer Signals

Background: Laser-Doppler Vibrometry (LDV) is a laser-based technique that allows measuring the motion of moving targets with high spatial and temporal resolution. To demonstrate its use for the measurement of carotid-femoral pulse wave velocity, a prototype system was employed in a clinical feasibility study. Data were acquired for analysis without prior quality control. Real-time application, however, will require a real-time assessment of signal quality. In this study, we (1) use template matching and matrix profile for assessing the quality of these previously acquired signals; (2) analyze the nature and achievable quality of acquired signals at the carotid and femoral measuring site; (3) explore models for automated classification of signal quality.Methods: Laser-Doppler Vibrometry data were acquired in 100 subjects (50M/50F) and consisted of 4–5 sequences of 20-s recordings of skin displacement, differentiated two times to yield acceleration. Each recording consisted of data from 12 laser beams, yielding 410 carotid-femoral and 407 carotid-carotid recordings. Data quality was visually assessed on a 1–5 scale, and a subset of best quality data was used to construct an acceleration template for both measuring sites. The time-varying cross-correlation of the acceleration signals with the template was computed. A quality metric constructed on several features of this template matching was derived. Next, the matrix-profile technique was applied to identify recurring features in the measured time series and derived a similar quality metric. The statistical distribution of the metrics, and their correlates with basic clinical data were assessed. Finally, logistic-regression-based classifiers were developed and their ability to automatically classify LDV-signal quality was assessed.Results: Automated quality metrics correlated well with visual scores. Signal quality was negatively correlated with BMI for femoral recordings but not for carotid recordings. Logistic regression models based on both methods yielded an accuracy of minimally 80% for our carotid and femoral recording data, reaching 87% for the femoral data.Conclusion: Both template matching and matrix profile were found suitable methods for automated grading of LDV signal quality and were able to generate a quality metric that was on par with the signal quality assessment of the expert. The classifiers, developed with both quality metrics, showed their potential for future real-time implementation.

Research on a Convolution Kernel Initialization Method for Speeding Up the Convergence of CNN

This paper presents a convolution kernel initialization method based on the local binary patterns (LBP) algorithm and sparse autoencoder. This method can be applied to the initialization of the convolution kernel in the convolutional neural network (CNN). The main function of the convolution kernel is to extract the local pattern of the image by template matching as the target feature of subsequent image recognition. In general, the Xavier initialization method and the He initialization method are used to initialize the convolution kernel. In this paper, firstly, some typical sample images were selected from the training set, and the LBP algorithm was applied to extract the texture information of the typical sample images. Then, the texture information was divided into several small blocks, and these blocks were input into the sparse autoencoder (SAE) for pre-training. After finishing the training, the weight values of the sparse autoencoder that met the statistical features of the data set were used as the initial value of the convolution kernel in the CNN. The experimental result indicates that the method proposed in this paper can speed up the convergence of the network in the network training process and improve the recognition rate of the network to an extent.

A high-performance brain switch based on code-modulated visual evoked potentials

Objective. Asynchronous brain-computer interfaces (BCIs) are more practical and natural compared to synchronous BCIs. A brain switch is a standard asynchronous BCI, which can automatically detect the specified change of the brain and discriminate between the control state and the idle state. The current brain switches still face challenges on relatively long reaction time (RT) and high false positive rate (FPR). Approach. In this paper, an online electroencephalography-based brain switch is designed to realize a fast reaction and keep long idle time (IDLE) without false positives (FPs) using code-modulated visual evoked potentials (c-VEPs). Two stimulation paradigms were designed and compared in the experiments: multi-code concatenate modulation (concatenation mode) and single-code periodic modulation (periodic mode). Using a task-related component analysis-based detection algorithm, EEG data can be decoded into a series of code indices. Brain states can be detected by a template matching approach with a sliding window on the output series. Main results. The online experiments achieved an average RT of 1.49 seconds when the average IDLE for each FP was 68.57 minutes (1.46e-2 FP/min) or an average RT of 1.67 seconds without FPs. Significance. This study provides a practical c-VEP based brain switch system with both fast reaction and low FPR during idle state, which can be used in various BCI applications.

The Saint-Ursanne earthquakes of 2000 revisited: evidence for active shallow thrust-faulting in the Jura fold-and-thrust belt

The interpretation of seismotectonic processes within the uppermost few kilometers of the Earth’s crust has proven challenging due to the often significant uncertainties in hypocenter locations and focal mechanisms of shallow seismicity. Here, we revisit the shallow seismic sequence of Saint-Ursanne of March and April 2000 and apply advanced seismological analyses to reduce these uncertainties. The sequence, consisting of five earthquakes of which the largest one reached a local magnitude (ML) of 3.2, occurred in the vicinity of two critical sites, the Mont Terri rock laboratory and Haute-Sorne, which is currently evaluated as a possible site for the development of a deep geothermal project. Template matching analysis for the period 2000–2021, including data from mini arrays installed in the region since 2014, suggests that the source of the 2000 sequence has not been persistently active ever since. Forward modelling of synthetic waveforms points to a very shallow source, between 0 and 1 km depth, and the focal mechanism analysis indicates a low-angle, NNW-dipping, thrust mechanism. These results combined with geological data suggest that the sequence is likely related to a backthrust fault located within the sedimentary cover and shed new light on the hosting lithology and source kinematics of the Saint-Ursanne sequence. Together with two other more recent shallow thrust faulting earthquakes near Grenchen and Neuchâtel in the north-central portion of the Jura fold-and-thrust belt (FTB), these new findings provide new insights into the present-day seismotectonic processes of the Jura FTB of northern Switzerland and suggest that the Jura FTB is still undergoing seismically active contraction at rates likely < 0.5 mm/yr. The shallow focal depths provide indications that this low-rate contraction in the NE portion of the Jura FTB is at least partly accommodated within the sedimentary cover and possibly decoupled from the basement.