Accurate and efficient lane detection is required for the effective operation of autonomous vehicles and advanced driver-assistance systems need it for their functioning. Traditional methods frequently use computationally demanding trigonometric operations, which are particularly challenging during line detection using the Hough Transform. One of the focused areas of this study is a real-time lane detection framework based on CORDIC technology. The CORDIC system operates as an iterative process that performs fundamental mathematical functions to enable its implementation on embedded platforms, whereas the matrix Mactor usage of CORDIC iterative methods presents an alternative to regular sine and cosine computations. The proposed pipeline implementation combines region-of-interest masking with Canny edge detection, modified Hough Transform, and CORDIC methods to detect multiple straight lane lines. The CORDIC implements the polar line equation using an iterative rotation method, thereby minimizing the computational requirements. The method can accurately extract multiple lane lines while significantly increasing the processing speed, as shown in the experimental results for both clear and rain-blurred highway images. The CORDIC-enhanced method shows advantages over standard algorithms through timing benchmarks, accumulator space visualizations, and performance metrics, which display the complete results of in-depth comparisons. This study demonstrates how hardware-oriented computation combined with algorithmic optimization enables real-time automotive applications and intelligent transportation systems to scale while achieving a correct rate of 98.72%.

Abstract This study proposes a machine vision-based automatic tool setting system for ultrasonic thread root rolling (UTRR), which addresses challenges in tool setting accuracy and the inefficiency of traditional manual or contact-based methods. The system integrates ROI extraction based on YOLOv5, an improved Canny edge detection algorithm, and the Iterative Least-Squares Piecewise Fitting Algorithm (ILS-PFA) to build a robust tool setting framework suitable for complex metallic threads under challenging conditions such as strong reflections and shadows. YOLOv5 is employed for reliable detection of the thread and rolling tool, even under varying lighting conditions. The improved Canny algorithm, augmented with adaptive bilateral filtering, Sobel gradients, and Otsu thresholding, effectively suppresses noise from metallic reflections and enhances edge clarity. The ILS-PFA method reconstructs thread profiles by fitting arc and line segments, while iterative optimization restores data lost due to shadows and reflections. A localization method based on curvature transition is used to precisely locate the thread root feature point, and the system calculates the spatial distance between the tool feature point and this feature point using camera calibration parameters. Experimental results demonstrate that the proposed method achieves micron-level measurement accuracy, with a mean absolute error of approximately 3.2 μm and a maximum relative error below 0.4% across different thread specifications. The tool-setting repeatability exhibits sub-micron stability, with a standard deviation below 0.6 μm. The total execution time of the automatic tool-setting process is approximately 1.78 s, satisfying practical industrial requirements. Error analysis indicates that the overall system error is controlled within 4 μm. These results confirm that the proposed method provides a robust, accurate, and cost-effective solution for automatic tool setting in UTRR applications.

With the rapid urbanization, the problem of municipal solid waste has become increasingly severe, and waste classification has become a key measure to achieve sustainable development. However, traditional manual sorting is inefficient and expensive, and the existing automatic identification methods based on deep learning have different limitations. In order to solve these problems, this paper proposes a waste classification scheme that combines Canny edge detection pretreatment with the lightweight ShuffleNet v2 network. First, denoise the image, and then use Canny edge detection to extract the waste profile to enhance the target characteristics and suppress background interference. Finally, enter the processed images into the ShuffleNet v2 network for classification. Experiments conducted on the kitchen garbage subset of Huawei’s garbage classification data set show that the proposed method achieves superior overall performance than SSD, YOLOv3 and unprocessed ShuffleNet v2. While maintaining the advantages of lightweight architecture, the method significantly improves the identification accuracy, thus expanding the technical path of garbage classification and accelerating its intelligent development.

Abstract Belt conveyors are widely utilized in material handling applications because of their high efficiency and substantial capacity. However, belt conveyor deviation typically leads to transmission system failures, which in turn impact production efficiency and may even cause serious safety incidents. Traditional sensor-based detection methods are sensitive to environmental noise, lighting changes, and material interference. Deep learning approaches face challenges such as high computational complexity, limited edge localization accuracy, and baseline drift due to camera position shifts. This paper proposes a deviation detection method that integrates deep learning with geometric constraints. The DC-YOLO-seg model, based on enhanced YOLOv11-seg, is combined with deformable convnets v3 (DCNv3) and coordinate attention (CA) mechanisms for high-precision instance segmentation of conveyor belts and rollers. Subsequently, the belt centerline was extracted using the Canny edge detection algorithm and random sample consensus (RANSAC) fitting method. The drive centerline was estimated based on the alignment of the roller centers, thereby quantifying the offset distance. This approach effectively reduces dependency on camera position and minimizes environmental interference. Experimental results on a self-constructed dataset demonstrate that DC-YOLO-seg achieves a Mask (mAP50-95) of 0.948, representing a 2.5% improvement over baseline models. The deviation detection error is generally maintained within 5 mm. This research provides a robust solution for intelligent operation and maintenance, establishing a foundation for cross-scenario generalization and real-time deployment.

Blind super-resolution (SR) aims to reconstruct high-resolution images from low-quality inputs under unknown degradation conditions. While numerous blind SR methods have been proposed in recent years, they still face critical limitations. Most approaches perform well under specific degradation patterns but struggle with complex scenarios involving multiple degradation factors and varying noise levels. This often leads to loss of structural integrity and fine details, resulting in suboptimal restoration quality. Furthermore, existing methods typically rely on convolutional neural networks (CNNs) with limited receptive fields, which hinders effective cross-domain information integration. Their inability to capture long-range dependencies compromises the reconstruction of global structures. In edge detection, conventional techniques frequently produce inaccurate or false edges, further degrading the quality of cross-domain integration and image restoration. To address these challenges, we propose Structural Integrity and Texture Fidelity Transformer (SITFFormer), a novel transformer-based framework for blind single-image SR. Our approach incorporates the Canny edge detection algorithm to accurately preserve true edges and suppress noise-induced artifacts, enhancing edge localization in complex and noisy environments. We also introduce the Cross-Domain Structure-Texture-Aware Network (CDSTNet), designed to integrate intra-domain and cross-domain features for comprehensive structure preservation and texture recovery. CDSTNet comprises two key modules: Cross-Domain Integration (CDI) that fuses intra- and cross-domain features to retain structural and textural details. Cross-Domain Learnable Attention (CDLA) that explores global dependencies, adaptively refines feature similarity, and filters out redundant non-local information. Both modules are equipped with a Cross-Attention Mechanism (CAM) to facilitate effective interaction and complementarity between domains, enhancing reconstruction fidelity. Extensive experiments on synthetic, noisy, and real-world datasets demonstrate that SITFFormer surpasses state-of-the-art methods in quantitative performance and visual quality, particularly in preserving structural integrity and recovering fine textures.

Insulators are the core components of high-voltage transmission lines, and the occurrence of low-/zero-value defects due to insulation degradation seriously endangers the safe and stable operation of power systems. Traditional detection methods for defective insulators suffer from low efficiency, poor anti-interference performance, and heavy reliance on manual experience, which are difficult to meet the demands of intelligent power grid operation and maintenance. In this work, a non-destructive detection method for low-/zero-value insulators based on infrared spectroscopy was proposed to realize the accurate and automatic identification and localization of defective insulators. Seven XP-100 porcelain suspension insulators were taken as the research objects, and an experimental platform was built in an artificial climate chamber to investigate the heating and discharge characteristics of zero-value insulators under different contamination conditions and spatial positions by combining infrared and ultraviolet imaging technologies, clarifying the typical low-temperature anomaly characteristic of zero-value insulators caused by volume resistance approaching zero. A two-dimensional axisymmetric model of insulator strings was constructed based on the finite element method, and the bidirectional coupling simulation of electric and temperature fields was carried out. The simulation results showed good consistency with the experimentally measured data, with a temperature difference error of less than 2 °C, which verified the reliability of the temperature field distribution law of defective insulators. An intelligent processing scheme for insulator infrared images was designed, which realized the automatic segmentation and feature extraction of insulator strings and key regions under complex backgrounds through Gaussian filtering denoising, histogram equalization, Hough transform angle correction, Canny edge detection and connected region analysis; it also extracted multidimensional features including temperature, texture and spatial distribution. Furthermore, an intelligent diagnostic model for low/zero-value insulators was established based on an improved back propagation (BP) neural network, which adopted a gradient descent optimization algorithm with momentum and adaptive learning rate to solve the problems of slow convergence and local minima of the traditional BP network. Taking corona discharge parameters (frequency, amplitude, and duration) and environmental temperature and humidity as input features, the model achieved automatic identification and precise positioning of low/zero-value insulators. Experimental results demonstrate that the proposed method effectively overcomes the limitations of traditional detection approaches, improves the accuracy and efficiency of defective insulator diagnosis, and realizes the integration of experimental mechanism analysis, multi-physics field simulation, intelligent image processing, and machine learning diagnosis. This research provides a reliable technical approach for the state monitoring of transmission line insulators and has important engineering application value for promoting the intellectualization of power equipment fault detection.

Canny edge detection for blurred textures in advertising art images based on generative adversarial networks

Magnetic particle imaging (MPI) is a functional imaging modality that enables highly sensitive tracking of magnetic nanoparticles. Field-free-line (FFL) MPI provides higher signal-to-noise ratio (SNR) than field-free-point MPI, however, noise in the sinogram domain and its propagation during reconstruction can introduce artifacts that degrade image quality. This study aims to develop a dual-domain denoising framework to improve SNR in both the sinogram and image domains for FFL-MPI tomographic reconstruction. We propose DudoMTD, a dual-domain cascaded network consisting of two components: (1) a Sinogram Denoising Transformer (SDT) integrates convolutional layers with a Vision Transformer to capture both local and long-range angular dependencies in the sinogram domain; and (2) an edge guiding autoencoder (EGA) operates in the image domain using convolutional filtering and an adaptive Canny operator to preserve structural boundaries. Simulated sinograms were generated using a standard FFL-MPI forward model based on the Langevin magnetization equation and system matrix formulation. The dataset consisted of 13039 simulated images, with 80% used for training and validation and 20% for testing. In addition, FFL-MPI phantom imaging data were used to evaluate the model under realistic measurement noise. Performance was compared with three benchmarks methods-DuDoNet, RED-CNN, and DnCNN-using peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and root mean square error (RMSE). Statistical significance was assessed using the Wilcoxon signed-rank test, with Benjamini-Hochberg correction for multiple comparisons. Effect sizes (Cohen's d) were reported to quantify the magnitude of improvements. Across four noise conditions (fixed SNRs of 15, 20, 30dB; 30dB with additional Poisson noise), DudoMTD showed statistically significant improvements (p<0.05) over all benchmark methods, with medium-to-large effect sizes. Specifically, the PSNR gains corresponded to Cohen's d values ranging from 0.378 to 1.311, while SSIM improvements yielded d values of 0.323-1.305. Performance gains were pronounced in the 20dB, 30dB, and 30dB + Poisson noise scenarios, where DudoMTD exceeded competing methods by 3%-10% in SSIM and demonstrated consistently superior structural preservation. In phantom experiments acquired on an in-house FFL-MPI system, DudoMTD achieved the highest average SNR (13.809) and effectively suppressed measured noise. DudoMTD mitigates dual-domain noise in FFL-MPI and improves tomographic image quality across diverse noise conditions. These improvements may facilitate downstream quantitative MPI applications, particularly in low-dose imaging scenarios.

LFM-Golay coding excitation anisotropy TFM with Canny operator second-order TGV for detecting deep defects in CFRP composites

Early and accurate detection of pigeon pea leaf diseases is essential for improving crop productivity and ensuring food security, particularly under real-field agricultural conditions. This paper introduces a shallow and computationally off-the-shelf deep learning system to detect the presence of pigeon pea leaf disease with great accuracy and in real-time on resource-limited cameras. DSLR and smartphone cameras were used to make up a custom high-resolution dataset under natural field conditions, including healthy leaves and major diseases, such as Fusarium wilt, leaf spot, and powdery mildew. All the images were downsampled to 224 × 224 pixels and processed with a Gaussian smoothing filter to remove noise and a Canny edge detector to improve structural features. Disease regions were accurately isolated using a Skill Optimization Algorithm (SOA)-driven segmentation strategy that dynamically optimized threshold levels, morphological kernel sizes, and lesion area constraints to handle background clutter and illumination variations. A pretrained EfficientNet-B0 model was used to extract deep semantic features, which consisted of compact 1280-dimensional feature vectors. A novel FMDDCN approach was used to classify these features through exploiting the sensitivity to subtle disease patterns by relying on differential feature modeling and multi-layer fusion of features. The model was fitted on stochastic gradient descent with a learning rate of 1 x 10-3 and a batch size of 32, and assessed on a 60/20/20 train validation test split with 5-fold cross-validation. The results of the experiment show consistent convergence with low overfitting. The proposed framework was found to produce a classification accuracy of 94.5%, precision of 91.0%, recall of 85.5% and Matthews Correlation Coefficient of 88.5% when it was used with four optimized features. In comparison, it is demonstrated that FMDDCN performs better than traditional machine learning and deep learning models, with its F1-score of 0.965 and the overall accuracy of 0.965. The suitability of the real-time edge deployment is verified, as confirmed by the use of computational analysis to reduce inference latency and memory consumption.

This paper provides an exploratory analysis of underwater video analysis techniques to enhance image quality and facilitate accurate classification of different marine species. Our methodology progresses through several steps, beginning with the quality of underwater images that might be reduced by variables such as decreased light intensity, color modification, and limited visibility. These attributes pose significant challenges to develop accurate object detection methods. This paper outlines the processing pipeline employed to enhance the quality of images from underwater videos and facilitate precise object detection. First, we use the Gray World (GW) algorithm for image enhancement, effectively mitigating the challenges of aquatic environment, such as color distortion and low contrast. Subsequently, we compare the traditional Histogram Equalization (HE) and the Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithms to assess their efficacy in enhancing underwater image quality. Next, Canny Edge Detection is utilized to identify the prominent features in the enhanced images, aiding in subsequent classification tasks. Next, three state-of-the-art deep learning models, Visual Geometry Group 16-layer network (VGG16), 50-layer Residual Network (ResNet50), and 121-layer Densely Connected Convolutional Network (DenseNet121), are leveraged through transfer learning to classify underwater species, including fish, coral reefs, and sea turtles. Finally, by enhancing the visual quality of underwater images, our research contributes to better understanding of the underwater ecosystem and supports conservation efforts. Enhanced Super-Resolution GAN (ESRGAN) is a superior Generative Adversarial Network (GAN) technique to improve the quality of noisy images. This paper contributes to advancing the field of underwater image and video analysis, offering valuable insights for applications in marine biology, environmental monitoring, underwater robotics, and autonomous navigation.

This study presents the development and field evaluation of an autonomous four-wheel drive (4WD) agricultural prototype equipped with a cost-effective image-based navigation system. While high-precision positioning typically relies on expensive RTK-GNSS systems, this research explores the operational limits of handcrafted feature extraction methods, specifically Canny Edge Detection and Probabilistic Hough Transform, on a resource-constrained Raspberry Pi 4B platform. The methodology includes structured field trials in a 30-metre corn field, with 10 successful autonomous runs conducted under three different lighting scenarios: sunny, cloudy, and twilight. Navigation accuracy was measured using Mean Cross Tracking Error (MCTE) with measurements recorded at 3-metre intervals. Results show that the system achieved its highest stability under cloudy (diffuse) conditions, with a minimum MCTE of 6.2 cm and 95% accuracy. A performance decrease was observed in twilight conditions (MCTE: 12.5 cm) due to a decrease in the signal-to-noise ratio (SNR) and in sunny conditions (MCTE: 8.0 cm) due to shadow-induced interference. The findings indicate that four-wheel drive platforms combined with optimised vision pipelines offer a viable, low-cost alternative for small-scale agricultural automation, provided that environmental lighting variability is addressed.

Background/Objectives: Accurate bone fracture detection is essential for orthopedic diagnosis and trauma management. Manual interpretation of X-ray or CT images can be time-consuming and may lead to inter-observer variability, particularly in subtle or multi-site fracture cases. This study proposes an interpretable Hybrid Selective Feature Network (Hybrid SFNet) to improve multi-site bone fracture detection performance and boundary localization. Methods: The proposed Hybrid SFNet extends the original SFNet architecture by incorporating multi-scale convolutional feature extraction and a semantic flow mechanism to enhance structural representation and fracture boundary delineation. Preprocessing techniques, including Canny edge detection, normalization, and data augmentation, were applied to improve feature quality. Model interpretability was addressed using Gradient-weighted Class Activation Mapping (Grad-CAM) to visualize regions contributing to predictions. The model was evaluated on publicly available multi-site fracture datasets using both standard and class-weighted loss configurations. Results: For binary fracture classification, the proposed model achieved 90 accuracy, 94% precision, 77% recall, and an F1-score of 85% for fractured cases. When class-weighted loss was applied, recall improved to 85%, reducing false negatives from 145 to 94 cases (approximately 35%). Under the weighted configuration, Cohen's Kappa reached 0.79 and the Matthews Correlation Coefficient (MCC) reached 0.76. Conclusions: The proposed Hybrid SFNet provides an interpretable and effective framework for multi-site bone fracture detection. The integration of multi-scale feature extraction and semantic flow mechanisms enhances detection performance and boundary localization, while Grad-CAM supports clinical interpretability. These results indicate the model's potential for supporting clinical decision-making in orthopedic imaging.

Batik is one of Indonesia’s cultural heritages, with motifs that are both diverse and intricate. The Kawung motif, characterized by repetitive circular patterns, is divided into sub-motifs such as Kawung Bribil, Kawung Sen, and Kawung Picis. Automatic classification of these sub-motifs is important for digital preservation but remains difficult due to subtle inter-class similarities. The aim of this research is to analyze the performance of VGG, ResNet, and DenseNet and determine the most effective CNN architecture in classifying the sub-motifs of Batik Kawung. The research method is a convolutional neural network-based image classification approach using a dataset of 300 Kawung Batik images evenly distributed across three classes. Preprocessing steps included grayscale conversion, resizing to 256 × 256 pixels, Canny edge detection, and normalization to the range [0,1]. The dataset was randomly split into 210 training, 60 validation, and 30 testing images. The results of this research are that VGG achieved the highest training accuracy of 97%, but only 67% on the testing set, indicating a tendency to overfit. In contrast, DenseNet achieved the best generalization performance with a testing accuracy of 80%, surpassing both VGG and ResNet. At the class level, DenseNet161 demonstrated consistent performance across all Kawung sub-motifs, with precision ranging from 67% to 91% and F1-scores between 71% and 95%. These results suggest that DenseNet161 not only performed effectively during training but also generalized well to unseen data, establishing it as the most robust architecture for sub-motif Batik Kawung classification. The results underscore the effectiveness of CNNs, particularly DenseNet, in classifying subtle batik sub-motifs. This research contributes to develope a reliable automated system for identifying Kawung batik, leveraging modern technology to support the preservation of Indonesia’s cultural heritage.

ABSTRACT Concrete structures are prone to degradation due to various internal flaws and external stresses, with crack formation being one of the most critical challenges affecting their strength and durability. Traditional methods of condition assessment are often limited by their inability to systematically detect and differentiate between crack types. In this study, a hybrid methodology is proposed in which manual crack assessment is complemented by classical image processing techniques, specifically, Otsu-Thresholding and Canny Edge Detection. Through this integration, the process of crack evaluation is automated and enhanced, allowing for more consistent identification and classification of cracks. The methodology is applied to real-world examples, where its effectiveness is demonstrated in detecting crack patterns at multiple scales and associating them with their underlying structural causes. It is shown that the proposed approach may provide a practical and resource-efficient tool for improving the consistency and reliability of structural assessments.

Tantalum’s supply chain instability demands efficient urban mining from e-waste. Here, we present an AI-enhanced process that combines intelligent sorting with sustainable hydrometallurgy for high-yield/high-purity Ta recovery. A hybrid sorting system, cascading an interpretable convolutional neural network (CNN) with automated multi-energy X-ray transmission (MEXRT) spectroscopy, achieved 99.6 % precision and 96.9 % recall at 3000 components/hour, resolving the Ta/Nb ambiguity. Spatial activation mapping illustrated the visual sorting mechanism, facilitating feature-driven upgrading. Meanwhile, Canny edge detection and K-edge detection enabled real-time and pixel-wise spectral analysis under multithreaded processing. Downstream, streamlined physical separation and thermodynamically guided reverse leaching selectively recovered Ta with 98.2 % efficiency under mild conditions. Advanced characterization using transmission electron microscopy and ion beam analysis revealed a quantifiable core-shell Ta/Ta 2 O 5 structure in leached products, guiding calcination into >99.8 % pure Ta 2 O 5 . This work establishes a closed-loop urban mining framework, demonstrating how AI and tailored refining enable a circular economy for critical metals.

The more years a ceramic plate has endured, the more cracks will form on its surface, creating a complex network. The interwoven cracks combine to form particles of various sizes and shapes. The identification and analysis of these particles can effectively determine the age of the plate, its material properties, the manufacturing techniques of that time, and its artistic value. This article presents a newly studied image segmentation algorithm for multiple crack delineation in ancient dish patterns. It consists of a special image preprocessing sub-algorithm that uses a newly designed fractional differential template different from the ordinary template, and it can remove noise more effectively; an improved Canny edge detector based on the image preprocessing, in which the two thresholds can be auto-decided according to the image information; and a number of post-functions for forming particles by cracks, which mainly rely on region splitting and merging based on particle shape analysis. The studied algorithm has been tested on more than a hundred ancient dish images, and it has been compared to several existing widely used image segmentation algorithms and methods, such as edge-detection-based, gray-level similarity-based, and deep learning methods; the results show that the new algorithm produces fewer over-segmentation and under-segmentation problems, and it works satisfactorily.

Optimised Canny edge detection algorithm for medical image feature mapping and extraction

Vision-based real-time monitoring and signal stabilization for automated lotus fiber extraction process

Abstract This study presents a real-time dynamic reconfiguration strategy for photovoltaic (PV) arrays under partial shading conditions, leveraging a metaheuristic Firebug Swarm Optimization (FSO) algorithm integrated with autonomous drone-based image analysis. High-resolution images of the PV array are captured throughout the day using an autonomous drone and subsequently processed using an Otsu–Canny edge detection algorithm to accurately remove noise, extract shading features, and classify the shading into distinct patterns. This processed shading information drives a Total Cross-Tied (TCT) switching matrix, which dynamically rewires a 5 × 5 PV array in real time without physically moving the modules. The proposed FSO-based reconfiguration framework optimizes both module-level connections and array power output, demonstrating a significant reduction in mismatch losses. Performance is rigorously evaluated across three distinct shading scenarios (10:00 AM, 1:00 PM, and 4:00 PM). Detailed analysis of Case 1 (Right-Bottom Diagonal shading) highlights the method’s effectiveness: the system with FSO-driven reconfiguration experimentally achieved 8265.68 W (with an efficiency of 15.2%), which is a substantial improvement over the 7599.24 W achieved without reconfiguration. Simulation results further validated this performance, reaching 9278.62 W. The proposed approach consistently outperforms conventional techniques like Grasshopper Optimization Algorithm (GOA), Grey Wolf Optimizer (GWO), and Particle Swarm Optimization (PSO) in terms of power yield, enhanced fill factor, and overall efficiency across all tested cases. The successful integration of drone-assisted shading detection, advanced image processing, and FSO-driven dynamic reconfiguration provides a robust, scalable, and efficient solution for maximizing PV energy yield under diverse and time-varying partial shading conditions, offering a clear implementation pathway for enhancing modern PV installations.