Manual Judgments Research Articles

Detection of black tea fermentation quality based on optimized deep neural network and hyperspectral imaging

The quality of black tea significantly relies on its fermentation process. Nevertheless, achieving precise and objective evaluations remains challenging due to the subjective nature of manual judgment involved in quality monitoring. To address this problem, hyperspectral imaging combined with the deep learning algorithms are proposed to identify the fermentation quality of black tea. Firstly, the hyperspectral data of Yinghong No. 9 black tea during five fermentation time intervals within 0–5 h are collected. Then, the Support Vector Machine (SVM), Artificial Neural Network (ANN), Partial Least Squares Discriminant Analysis (PLS-DA), and Naive Bayesian (NB) are used to construct black tea fermentation quality detection models based on full spectrum and selected spectrum data. Furthermore, deep learning algorithms including the Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), CNN-LSTM, and Swarm Optimization (PSO) optimized CNN-LSTM (PSO-CNN-LSTM) are also used to build the detection model using the spectral images. The experimental results indicate that deep learning algorithms have obvious advantages over traditional machine learning algorithms in tea fermentation quality detection. Besides, the PSO-CNN-LSTM model shows the best classification performance compared to other algorithms and achieves an accuracy of 96.78% on the test set. This study demonstrates the significant potential of combining deep learning with hyperspectral imaging for predicting black tea fermentation quality. This provides a new approach for effective monitoring of the black tea fermentation process and a useful reference for other applications in similar fields.

Investigation, Evaluation, and Dynamic Monitoring of Traditional Chinese Village Buildings Based on Unmanned Aerial Vehicle Images and Deep Learning Methods

The investigation, evaluation, and dynamic monitoring of traditional village buildings are crucial for the protection and inheritance of their architectural styles. This study takes traditional villages in Shandong Province, China, as an example, employing UAV images and deep learning technology. Utilizing the YOLOv8 instance segmentation model, it introduces three key features reflecting the condition of traditional village buildings: roof status, roof form, and courtyard vegetation coverage. By extracting feature data on the condition of traditional village buildings and constructing a transition matrix for building condition changes, combined with corresponding manual judgment assistance, the study classifies, counts, and visually outputs the conditions and changes of buildings. This approach enables the investigation, evaluation, and dynamic monitoring of traditional village buildings. The results show that deep learning technology significantly enhances the efficiency and accuracy of traditional village architectural investigation and evaluations, and it performs well in dynamic monitoring of building condition changes. The “UAV image + deep learning” technical system, with its simplicity, accuracy, efficiency, and low cost, can provide further data and technical support for the planning, protection supervision, and development strategy formulation of traditional Chinese villages.

Automated localization of dike leakage outlets using UAV-borne thermography and YOLO-based object detectors

Leakage-induced soil erosion poses a major threat to dike failure, particularly during floods. Timely detection and notification of leakage outlets to dike management are crucial for ensuring dike safety. However, manual inspection, the current main approach for identifying leakage outlets, is costly, inefficient, and lacks spatial coverage. To achieve efficient and automatic localization of dike leakage outlets, an innovative strategy combining drones, infrared thermography, and deep learning is presented. Drones are employed for dikes’ surface sensing. Real-time images from these drones are sent to a server where well-trained detectors are deployed. Once a leakage outlet is detected, alarming information is remotely sent to dike managers. To realize this strategy, 4 thermal imagers were employed to image leaking outlets of several models and actual dikes. 9,231 hand-labeled thermal images with 13,387 leaking objects were selected for analysis. 19 detectors were trained using transfer learning. The best detector achieved a mean average precision of 95.8 % on the challenging test set. A full-scale embankment was constructed for leakage outlet detection tests. Various field tests confirmed the efficiency of the proposed leakage outlet localization method. In some tough conditions, the trained detector also evidently outperformed manual judgement. Results indicate that under typical circumstances, the localization error of the proposed method is within 5 m, demonstrating its practical reliability. Finally, the influencing factors and limits of the suggested strategy are thoroughly examined.

Category Mapping of Emergency Supplies Classification Standard Based on BERT-TextCNN

In recent years, the escalation in emergency occurrences has underscored the pressing need for expedient responses in delivering essential supplies. Efficient integration and precise allocation of emergency resources under joint government–enterprise stockpiling models are pivotal for enhancing emergency response effectiveness and minimizing economic repercussions. However, current research predominantly focuses on contract coordination and cost-sharing within these joint reserve modes, overlooking significant discrepancies in emergency supply classification standards between government and enterprise sectors, as well as the asymmetry in cross-sectoral and cross-regional supply information. This oversight critically impedes the timeliness and accuracy of emergency supply responses. In practice, manual judgment has been used to match the same materials under differing classification standards between government and enterprise reserves. Still, this approach is inefficient and prone to high error rates. To mitigate these challenges, this study proposes a methodology leveraging the BERT pre-trained language model and TextCNN neural network to establish a robust mapping relationship between these classification criteria. The approach involves abstracting textual representations of both taxonomical classes, generating comparable sentence vectors via average pooling, and calculating cosine similarity scores to facilitate precise classification mapping. Illustrated with China’s Classification and Coding of Emergency Supplies standards and Global Product Classification standards, empirical validation on annotated data demonstrates the BERT-TextCNN model’s exceptional accuracy of 98.22%, surpassing other neural network methodologies such as BERT-CNN, BERT-RNN, BERT-BiLSTM, etc. This underscores the potential of advanced neural network techniques in enhancing emergency supply management across diverse sectors and regions.

Research on the Method for Recognizing Bulk Grain-Loading Status Based on LiDAR.

Grain is a common bulk cargo. To ensure optimal utilization of transportation space and prevent overflow accidents, it is necessary to observe the grain's shape and determine the loading status during the loading process. Traditional methods often rely on manual judgment, which results in high labor intensity, poor safety, and low loading efficiency. Therefore, this paper proposes a method for recognizing the bulk grain-loading status based on Light Detection and Ranging (LiDAR). This method uses LiDAR to obtain point cloud data and constructs a deep learning network to perform target recognition and component segmentation on loading vehicles, extract vehicle positions and grain shapes, and recognize and make known the bulk grain-loading status. Based on the measured point cloud data of bulk grain loading, in the point cloud-classification task, the overall accuracy is 97.9% and the mean accuracy is 98.1%. In the vehicle component-segmentation task, the overall accuracy is 99.1% and the Mean Intersection over Union is 96.6%. The results indicate that the method has reliable performance in the research tasks of extracting vehicle positions, detecting grain shapes, and recognizing loading status.

Text classification for distribution substation inspection based on BERT-TextRCNN model

With the advancement of source-load interaction in the new power systems, data-driven approaches have provided a foundational support for aggregating and interacting between sources and loads. However, with the widespread integration of distributed energy resources, fine-grained perception of intelligent sensing devices, and the inherent stochasticity of source-load dynamics, a massive amount of raw data is being recorded and accumulated in the data center. Valuable information is often dispersed across different paragraphs of the raw data, making it challenging to extract effectively. Distribution substation inspection plays a crucial role in ensuring the safe operation of the power system. Traditional methods for inspection report text classification typically rely on manual judgment and accumulated experience, resulting in low efficiency and a significant misjudgment rate. Therefore, this paper proposes a text classification method for inspection reports based on the pre-trained BERT-TextRCNN model. By utilizing the dense connection between the BERT embedding layer and the neural network, the proposed method improves the accuracy of matching long texts. This article collected 2,831 maintenance data for the first quarter of 2023 from the distribution room, including approximately 58 environmental testing data, 738 environmental box testing data, approximately 672 distribution room testing data, and approximately 1,363 box type substation testing data. A text corpus was constructed for experiments. Experimental results demonstrate that the proposed model automatically classifies a large volume of manually recorded inspection report data based on time, location, and faults, achieving a classification accuracy of 94.7%, precision of 92%, recall of 92%, and F1 score of 90.3%.

Research on wavelet packet energy entropy extraction method for acoustic signal of tunnel lining cavity

Abstract Tunnel lining cavity is a common disease in high-speed railway tunnels, which seriously affects the safe operation of the line.At present, the commonly used method is manual knocking inspection, but it relies on manual judgment, subjective factors have great influence, and there are errors. This paper proposes a feature extraction method of acoustic vibration signal of tunnel lining cavity based on wavelet packet energy entropy and uses wavelet packet energy entropy to realize the identification of cavity defects.The full-scale simulation model of tunnel with a radius of 6m was constructed by COMSOL and different cavity conditions were set up. The wavelet packet energy entropy characteristics of acoustic signals were extracted and analyzed. The mapping relationship between cavity depth and cavity size and wavelet packet energy entropy was explored. The wavelet packet energy entropy and frequency band distribution characteristics were extracted to realize the identification of cavity defects.The local model which can characterize the whole tunnel is designed and the embedded defects are verified by experiments. The results show that the wavelet packet energy entropy can be used as the identification feature of tunnel lining cavity defects. When the wavelet packet energy entropy is less than 2.53, the existence of tunnel lining cavity defects is characterized.When the wavelet packet energy entropy is in the range of 2.53-2.94, it characterizes the existence of the transition boundary between tunnel lining cavity defect and no disease.This index can be used to realize the rapid and effective identification of cavity defects.

Augmented Reality and Artificial Intelligence Medical Waste Classification System and Method

There are four categories of medical waste that cannot be mixed as this will cause serious problems suchas environmental pollution or infection. In the past, the classification of medical waste often involved a lot ofhuman and material resources to process, with workers at risk of exposure to infectious substances. Therefore, an augmented reality (AR) and artificial intelligence (AI) medical waste classification system and method were developed. This innovative medical waste classification system and method combines AR and AI identification technology to reduce the risk of manual judgment errors by clinical staff when handling medical waste.

A multiclass deep learning algorithm for healthy lung, Covid-19 and pneumonia disease detection from chest X-ray images

A crucial step in the battle against the coronavirus disease 2019 (Covid-19) pandemic is efficient screening of the Covid affected patients. Deep learning models are used to improve the manual judgements made by healthcare professionals in classifying Chest X-Ray (CXR) images into Covid pneumonia, other viral/bacterial pneumonia, and normal images. This work uses two open source CXR image dataset having a total of 15,153 (dataset 1), and 4575 (dataset 2) images respectively. We trained three neural network models with a balanced subset of dataset 1 (1345 images per class), balanced dataset 2 (1525 images per class), and an unbalanced full dataset 1. The models used are VGG16 and Inception Resnet (IR) using transfer learning and a tailor made Convolutional Neural Network (CNN). The first model, VGG16 gives an accuracy, sensitivity, specificity, and F1 score of 96%, 97.8%, 95.92%, 97% respectively. The second model, IR gives an accuracy, sensitivity, specificity and F1 score of 97%, 98.51%, 97.28%, 99% respectively. The third and best proposed model, CNN gives an accuracy, sensitivity, specificity, and F1 score of 97%, 98.21%, 96.62%, 98% respectively. These performance metrics were obtained for the balanced dataset 1 and all models used 80:10:10 cross validation technique. The highest accuracy using CNN for all the three datasets are 97%, 96%, and 93% respectively. Gradient-weighted Class Activation Mapping (Grad-CAM) is used to ensure that the model uses genuine pathology markers to generalize.

Working condition recognition of fused magnesium furnace based on stochastic configuration networks and reinforcement learning

Automatic and accurate recognition of abnormal working conditions of fused magnesia furnace is of great significance to the safe and reliable production of fused magnesia. Aiming at the defects of manual judgment of abnormal working conditions in the production process of fused magnesium furnace and the existing recognition method of abnormal working conditions based on machine learning, this paper proposes a working condition recognition model for fused magnesium furnace based on stochastic configuration networks and reinforcement learning. Firstly, a hybrid data augmentation method of generative and non-generative is used to obtain high-quality sample data with salient features. Secondly, based on ResNeXt, multi-scale local features are extracted under the condition of limited parameter quantity by controlling the cardinality. Combining a mixed model that enjoys the benefit of both self-attention and convolution (ACmix) and bidirectional feature pyramid network (BiFPN), the extracted local feature maps of different scales are cross-scale fused and focused, and more differentiated detailed feature information of the region of interest is retained. Thirdly, based on Transformer, the working condition recognition network of fused magnesium furnace is constructed to improve the global correlation between adjacent local features in the spatial dimension. The fused features are sent to stochastic configuration networks to establish a classification criterion for working condition recognition of fused magnesium furnace with generalization ability. Finally, reinforcement learning is used to evaluate the credibility of uncertain recognition results of samples in real time, and a self-optimizing adjustment action strategy at the Transformer encoding layer is defined. Build a library of Transformer models with different encoding layers, which is adapt to the different feature extraction requirements of multi-modal working samples. The experimental results show that the method in this paper has better recognition performance and generalization ability than other algorithms.

Microscopic identification of foodborne bacterial pathogens based on deep learning method

Accurate and rapid detection of foodborne bacterial pathogens is critical for food quality control. Nowadays, tracking morphological bacterial properties using microscope is still a priority at the grass-roots food supervision department due to its simplicity and low cost. However, the method requires highly professional personnel and there are certain misjudgments in the process of analysis. Automatically recognizing foodborne pathogen using deep learning algorithm to replace manual microscopy will not only reduce expert cost, artificial misjudgment, and operation time in detection, but also provide more objective and accurate identification. Here, we firstly constructed a high-quality and large-scale dataset of foodborne pathogenic bacteria, allowing the deep learning-based model to be efficiently trained and achieve accurate identification. The deep convolutional neural network-based model is capable of identifying six common foodborne pathogens, including Escherichia coli (O157:H7), Vibrio parahaemolyticus, Staphylococcus aureus, Bacillus cereus, Salmonella typhi, and Streptococcus hemolyticus, with accuracy rates of 90%–100%. This method can assist or replace the manual microscopic inspection step in traditional detection methods, and is promising to break through the traditional approach that heavily relies on manual judgment, greatly reduce the cost of experts and human errors, and provide rapid, accurate, and powerful discriminatory results in large quantities for detection.

Automated identification and segmentation of urine spots based on deep-learning.

Micturition serves an essential physiological function that allows the body to eliminate metabolic wastes and maintain water-electrolyte balance. The urine spot assay (VSA), as a simple and economical assay, has been widely used in the study of micturition behavior in rodents. However, the traditional VSA method relies on manual judgment, introduces subjective errors, faces difficulty in obtaining appearance time of each urine spot, and struggles with quantitative analysis of overlapping spots. To address these challenges, we developed a deep learning-based approach for the automatic identification and segmentation of urine spots. Our system employs a target detection network to efficiently detect each urine spot and utilizes an instance segmentation network to achieve precise segmentation of overlapping urine spots. Compared with the traditional VSA method, our system achieves automated detection of urine spot area of micturition in rodents, greatly reducing subjective errors. It accurately determines the urination time of each spot and effectively quantifies the overlapping spots. This study enables high-throughput and precise urine spot detection, providing important technical support for the analysis of urination behavior and the study of the neural mechanism underlying urination.

U-fiber analysis: a toolbox for automated quantification of U-fibers and white matter hyperintensities.

Whether white matter hyperintensities (WMHs) involve U-fibers is of great value in understanding the different etiologies of cerebral white matter (WM) lesions. However, clinical practice currently relies only on the naked eye to determine whether WMHs are in the vicinity of U-fibers, and there is a lack of good neuroimaging tools to quantify WMHs and U-fibers. Here, we developed a multimodal neuroimaging toolbox named U-fiber analysis (UFA) that can automatically extract WMHs and quantitatively characterize the volume and number of WMHs in different brain regions. In addition, we proposed an anatomically constrained U-fiber tracking scheme and quantitatively characterized the microstructure diffusion properties, fiber length, and number of U-fibers in different brain regions to help clinicians to quantitatively determine whether WMHs in the proximal cortex disrupt the microstructure of U-fibers. To validate the utility of the UFA toolbox, we analyzed the neuroimaging data from 246 patients with cerebral small vessel disease (cSVD) enrolled at Zhongshan Hospital between March 2018 and November 2019 in a cross-sectional study. According to the manual judgment of the clinician, the patients with cSVD were divided into a WMHs involved U-fiber group (U-fiber-involved group, 51 cases) and WMHs not involved U-fiber group (U-fiber-spared group, 163 cases). There were no significant differences between the U-fiber-spared group and the U-fiber-involved group in terms of age (P=0.143), gender (P=0.462), education (P=0.151), Mini-Mental State Examination (MMSE) scores (P=0.151), and Montreal Cognitive Assessment (MoCA) scores (P=0.411). However, patients in the U-fiber-involved group had higher Fazekas scores (P<0.001) and significantly higher whole brain WMHs (P=0.046) and deep WMH volumes (P<0.001) compared to patients in the U-fiber-spared group. Moreover, the U-fiber-involved group had higher WMH volumes in the bilateral frontal [P(left) <0.001, P(right) <0.001] and parietal lobes [P(left) <0.001, P(right) <0.001]. On the other hand, patients in the U-fiber-involved group had higher mean diffusivity (MD) and axial diffusivity (AD) in the bilateral parietal [P(left, MD) =0.048, P(right, MD) =0.045, P(left, AD) =0.015, P(right, AD) =0.015] and right frontal-parietal regions [P(MD) =0.048, P(AD) =0.027], and had significantly reduced mean fiber length and number in the right parietal [P(length) =0.013, P(number) =0.028] and right frontal-parietal regions [P(length) =0.048] compared to patients in the U-fiber-spared group. Our results suggest that WMHs in the proximal cortex may disrupt the microstructure of U-fibers. Our tool may provide new insights into the understanding of WM lesions of different etiologies in the brain.

Semantic analysis and construction of English discourse based on neural network

Abstract Accurate recognition and analysis of English semantics is of great significance in the application field of computer and artificial intelligence technology. Traditional text analysis technology has poor real-time analysis accuracy and efficiency. This paper constructs an English semantic analysis model based on neural networks. By designing an English semantic framework that integrates LSTM and RNN and optimizing it on the basis of LSTM and RNN models, an English semantic analysis model with improved attention mechanisms and Bi-LSTM is constructed. By comparing the Bi-LSTM+Attention model proposed in this paper with other semantic analysis models, the classification accuracy of the model constructed in this paper reaches 98%. Through the practical application of the model to analyze the semantic coherence of English, the average score of the semantic coherence quality measured by the model is 15.76, and the average score of the manual judgment is 16.03, which indicates that the application of the semantic analysis model constructed in this paper is close to the effect of the manual judgment.

Analysis of influencing factors of urban residents' physical health based on logistic regression method

This article aims to conduct an in-depth study of residents' eating habits, lifestyles and their close connection with physical health to provide a strong scientific basis to deal with the increasing problem of chronic diseases in society. In order to solve this problem, this article uses a variety of sophisticated data analysis methods and models to explore the following core issues. Firstly, the credibility and validity of the research data were ensured by conducting reliability and validity tests on the questionnaire data; secondly, in the data processing, manual judgment and statistical principle judgment were used to identify and process outliers, thereby The quality of the data was improved; then, the logistic regression model was used to analyze the relationship and degree of correlation between the hypertension prevalence data and various indicators, and to explore in depth whether these factors were significantly related to the occurrence of chronic diseases. Finally, by comparing the research results, this article summarizes the connection between diet structure, lifestyle and other related factors and common chronic diseases, providing strong support for improving the health status of residents, which is an important contribution of this study.

Research Review on Quality Detection of Fresh Tea Leaves Based on Spectral Technology.

As the raw material for tea making, the quality of tea leaves directly affects the quality of finished tea. The quality of fresh tea leaves is mainly assessed by manual judgment or physical and chemical testing of the content of internal components. Physical and chemical methods are more mature, and the test results are more accurate and objective, but traditional chemical methods for measuring the biochemical indexes of tea leaves are time-consuming, labor-costly, complicated, and destructive. With the rapid development of imaging and spectroscopic technology, spectroscopic technology as an emerging technology has been widely used in rapid non-destructive testing of the quality and safety of agricultural products. Due to the existence of spectral information with a low signal-to-noise ratio, high information redundancy, and strong autocorrelation, scholars have conducted a series of studies on spectral data preprocessing. The correlation between spectral data and target data is improved by smoothing noise reduction, correction, extraction of feature bands, and so on, to construct a stable, highly accurate estimation or discrimination model with strong generalization ability. There have been more research papers published on spectroscopic techniques to detect the quality of tea fresh leaves. This study summarizes the principles, analytical methods, and applications of Hyperspectral imaging (HSI) in the nondestructive testing of the quality and safety of fresh tea leaves for the purpose of tracking the latest research advances at home and abroad. At the same time, the principles and applications of other spectroscopic techniques including Near-infrared spectroscopy (NIRS), Mid-infrared spectroscopy (MIRS), Raman spectroscopy (RS), and other spectroscopic techniques for non-destructive testing of quality and safety of fresh tea leaves are also briefly introduced. Finally, in terms of technical obstacles and practical applications, the challenges and development trends of spectral analysis technology in the nondestructive assessment of tea leaf quality are examined.

A systematic framework for tackling anomalous pre-welding workpiece postures with regular butt joints based on prototype features

In intelligent welding systems, pre-welding parameter extraction is a foremost technology in upgrading robotic welding and integrating emerging technologies, e.g., digital twins, big data, and cloud manufacturing. However, current workpiece postures still rely on manual judgments based on workers' experience, which has become one of major issues hindering the further advancement of intelligent welding systems towards mass production. To cope with this issue, aiming at four typical regular butt joints, a systematic tackling framework is proposed and carried out from posture description, posture feature construction, posture metrics to visualization model reconstruction. At a core, the prototype feature is proposed to characterize pre-welding workpiece postures and a series of text characters based on it is introduced to describe various anomalous workpiece postures including concurrent tilt, misalignment, seam variation, and stacking between them. A comprehensive process for constructing prototype features is performed from data acquisition, image processing to key point search, among which the algorithms for extracting differential features of different seam types are integrated. Based on the constructed prototype features, several posture metric parameters are defined, and workpiece posture models can be easily reconstructed. In addition, the good generalizability of the proposed framework for seam types with regular edge and seam features is also discussed. Ultimately, experimental results show that the prototype feature-based posture description of the pre-welding workpiece can efficiently and accurately characterize multiple anomalous postures.

An efficient approach to detect and track winter flush growth of litchi tree based on UAV remote sensing and semantic segmentation

The immature winter flush affects the flower bud differentiation, flowering and fruit of litchi, and then seriously reduces the yield of litchi. However, at present, the area estimation and growth process monitoring of winter flush still rely on manual judgment and operation, so it is impossible to accurately and effectively control flush. An efficient approach is proposed in this paper to detect the litchi flush from the unmanned aerial vehicle (UAV) remoting images of litchi crown and track winter flush growth of litchi tree. The proposed model is constructed based on U-Net network, of which the encoder is replaced by MobeilNetV3 backbone network to reduce model parameters and computation. Moreover, Convolutional Block Attention Module (CBAM) is integrated and convolutional layer is added to enhance feature extraction ability, and transfer learning is adopted to solve the problem of small data volume. As a result, the Mean Pixel Accuracy (MPA) and Mean Intersection over Union (MIoU) on the flush dataset are increased from 90.95% and 83.3% to 93.4% and 85%, respectively. Moreover, the size of the proposed model is reduced by 15% from the original model. In addition, the segmentation model is applied to the tracking of winter flushes on the canopy of litchi trees and investigating the two growth processes of litchi flushes (late-autumn shoots growing into flushes and flushes growing into mature leaves). It is revealed that the growth processes of flushes in a particular branch region can be quantitatively analysed based on the UAV images and the proposed semantic segmentation model. The results also demonstrate that a sudden drop in temperature can promote the rapid transformation of late-autumn shoots into flushes. The method proposed in this paper provide a new technique for accurate management of litchi flush and a possibility for the area estimation and growth process monitoring of winter flush, which can assist in the control operation and yield prediction of litchi orchards.

Automatic defect detection in operational high-speed railway tunnels guided by train-mounted ground penetrating radar data

The tunnel diseases of high-speed railways present significant risks under the substantial dynamic loads from trains. Train-mounted ground-penetrating radar detection data for tunnels relies heavily on manual judgment, resulting in low efficiency and inaccurate identification. Meanwhile, automatic detection models need more samples. We propose an automated recognition framework, guided by generated data, to address these issues for identifying defects in operational high-speed railway tunnels. In this work, we propose a multi-category GPR generative adversarial network (MGPR-GAN) to generate many defect data across different categories to guide YOLOv5 training instead of relying predominantly on real data. It then uses generated data (65.43%) alongside a smaller proportion of real defect data (34.57%) to guide a YOLOv5 model. The model outputs the defect category, confidence level, and defect width and height estimates in conjunction with prior information. We validated our method using samples from 71 tunnels. The results reveal that our proposed method outperforms the average accuracy of the three advanced identification methods (FasterRCNN, SSD, and YOLOv5) by over 20.62%.

Application of Task-Aligned Model Based on Defect Detection

In recent years, with the rise of the automation wave, reducing manual judgment, especially in defect detection in factories, has become crucial. The automation of image recognition has emerged as a significant challenge. However, the problem of how to effectively improve the classification of defect detection and the accuracy of the mean average precision (mAP) is a continuous process of improvement and has evolved from the original visual inspection of defects to the present deep learning detection system. This paper presents an application of deep learning, and the task-aligned approach is firstly used on metal defects, and the anchor and bounding box of objects and categories are continuously optimized by mutual correction. We used the task-aligned one-stage object detection (TOOD) model, then improved and optimized it, followed by deformable ConvNets v2 (DCNv2) to adjust the deformable convolution, and finally used soft efficient non-maximum suppression (Soft-NMS) to optimize intersection over union (IoU) and adjust the IoU threshold and many other experiments. In the Northeastern University surface defect detection dataset (NEU-DET) for surface defect detection, mAP increased from 75.4% to 77.9%, a 2.5% increase in mAP, and mAP was also improved compared to existing advanced models, which has potential for future use.