Pixel Values Research Articles

A chaotic based image encryption scheme using elliptic curve cryptography and genetic algorithm

In the era of digital communication and data security, image encryption plays a crucial role in safeguarding sensitive information. Protecting sensitive visual data from unauthorized access drives the pursuit of advanced image encryption methods. This paper proposes a novel approach to enhance image encryption by combining the power of a chaotic map, elliptic curve cryptography, and genetic algorithm. The chaotic map, specifically Arnold’s cat map, is employed to introduce chaos and randomness into the encryption process. The proposed image encryption process involves applying Arnold’s cat map for shuffling the pixel positions, followed by elliptic curve cryptography for encrypting the pixel values using public and private keys. Additionally, a genetic algorithm is employed to optimize the key generation process, enhancing the security of the encryption scheme. The combined utilization of these techniques aims to achieve a high level of confidentiality and robustness in image encryption. The algorithm underwent thorough analysis. It achieved a maximum entropy score of 7.99, indicating a high level of randomness and unpredictability in the encrypted data. Additionally, it exhibited near-zero correlation, which suggests strong resistance against statistical attacks. Moreover, the cryptographic range of possible keys was found to be 2511\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$2^{511}$$\\end{document}. This extensive key space makes the algorithm highly resilient against brute force attacks. It took only 0.5634 s to encrypt a moderately sized 512×512\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$512\ imes 512$$\\end{document} pixel image with an 8-bit image on a standard desktop computer with a 2.3 GHz processor and 16 GB of RAM. The experimental findings confirm that the proposed approach is highly effective and efficient in safeguarding sensitive image data from unauthorized access and potential attacks. This scheme has the benefit of allowing us to protect our private image data while it’s being transmitted.

Transfer Learning with ResNet3D-101 for Global Prediction of High Aerosol Concentrations

In order to better predict the high aerosol concentrations associated with air pollution and climate change, a machine learning model was developed using transfer learning and the segmentation process of global satellite images. The main concept of transfer learning lies on convolutional neural networks and works by initializing the already trained model weights to better adapt the weights when the network is trained on a different dataset. The transfer learning technique was tested with the ResNet3D-101 model pre-trained from a 2D ImageNet dataset. This model has performed well for contrail detection to assess climate impact. Aerosol distributions can be monitored via satellite remote sensing. Satellites can monitor some aerosol optical properties like aerosol optical thickness. Aerosol optical thickness snapshots were the input dataset for the model and were obtained from NASA’s Terra-Modis satellite; the output images were segmented by comparing the pixel values with a threshold value of 0.8 for aerosol optical thickness. Hyperparameter optimization finds a tuple of hyperparameters that yields an optimal model that minimizes a predefined loss function on given independent data. The model structure was adjusted in order to improve the performance of the model by applying methods and hyperparameter optimization techniques such as grid search, batch size, threshold, and input length. According to the criteria defined by the authors, the distance domain criterion and time domain criterion, the developed model is capable of generating adequate data and finding patterns in the time domain. As observed from the comparison of relative coefficients for the criteria metrics proposed by the authors, ddc and dtc, the deep learning model based on ConvLSTM layers developed in our previous studies has better performance than the model developed in this study with transfer learning.

Enhancing Diagnostic Images to Improve the Performance of the Segment Anything Model in Medical Image Segmentation.

Medical imaging serves as a crucial tool in current cancer diagnosis. However, the quality of medical images is often compromised to minimize the potential risks associated with patient image acquisition. Computer-aided diagnosis systems have made significant advancements in recent years. These systems utilize computer algorithms to identify abnormal features in medical images, assisting radiologists in improving diagnostic accuracy and achieving consistency in image and disease interpretation. Importantly, the quality of medical images, as the target data, determines the achievable level of performance by artificial intelligence algorithms. However, the pixel value range of medical images differs from that of the digital images typically processed via artificial intelligence algorithms, and blindly incorporating such data for training can result in suboptimal algorithm performance. In this study, we propose a medical image-enhancement scheme that integrates generic digital image processing and medical image processing modules. This scheme aims to enhance medical image data by endowing them with high-contrast and smooth characteristics. We conducted experimental testing to demonstrate the effectiveness of this scheme in improving the performance of a medical image segmentation algorithm.

TGLFusion: A Temperature-Guided Lightweight Fusion Method for Infrared and Visible Images.

The fusion of infrared and visible images is a well-researched task in computer vision. These fusion methods create fused images replacing the manual observation of single sensor image, often deployed on edge devices for real-time processing. However, there is an issue of information imbalance between infrared and visible images. Existing methods often fail to emphasize temperature and edge texture information, potentially leading to misinterpretations. Moreover, these methods are computationally complex, and challenging for edge device adaptation. This paper proposes a method that calculates the distribution proportion of infrared pixel values, allocating fusion weights to adaptively highlight key information. It introduces a weight allocation mechanism and MobileBlock with a multispectral information complementary module, innovations which strengthened the model's fusion capabilities, made it more lightweight, and ensured information compensation. Training involves a temperature-color-perception loss function, enabling adaptive weight allocation based on image pair information. Experimental results show superiority over mainstream fusion methods, particularly in the electric power equipment scene and publicly available datasets.

Quantum image encryption algorithm based on four-dimensional chaos

Background: Quantum image processing is rapidly developing in the field of quantum computing, and it can be successfully implemented on the Noisy Intermediate-Scale Quantum (NISQ) device. Quantum image encryption holds a pivotal position in this domain. However, the encryption process often encounters security vulnerabilities and entails complex computational complexities, thereby consuming substantial quantum resources. To address this, the present study proposes a quantum image encryption algorithm based on four-dimensional chaos.Methods: The classical image is first encoded into quantum information using the Generalized Quantum Image Representation (GQIR) method. Subsequently, the trajectory of the four-dimensional chaotic system is randomized, and multi-dimensional chaotic keys are generated to initially encrypt the pixel values of the image. Then, the Arnold transformation is applied to randomly encrypt the pixel positions, resulting in the encrypted image. During the decryption process, the inverse process of encryption is employed to restore the original image.Results: We simulated this process in the Python environment, and the information entropy analysis experiment showed that the information entropy of the three encrypted images reached above 7.999, so the system has good encryption. At the same time, the correlation of the pixel distribution after the encryption algorithm is weak, which proves that the control parameters of the chaotic system can effectively reduce the correlation between pixels in the image. In the final key space analysis, the key space issued by our encryption can reach $10140\gg 2128$.Conclusion: Our method is resistant to destructive attacks and can produce scrambled images with higher encryption and usability. This algorithm solves the problems of general encryption algorithms such as periodicity, small key space, and vulnerability to statistical analysis, and proposes a reliable and effective encryption scheme. By making full use of the characteristics of Arnold transformation permutation, ergodicity and the randomness of the four-dimensional chaotic system, the encryption algorithm uses the larger key space provided by the four-dimensional Lorenz system.

Cross-channel color image encryption through 2D hyperchaotic hybrid map of optimization test functions

The security of chaos-based Image Encryption (IE) algorithms inherently depends on the permutation and diffusion strategy of the algorithm and the dynamic performance of chaotic maps. However, the existing studies suffer from low disordering capability, poor dynamic performance, and lack of sensitivity of the chaotic maps. In order to address these limitations, a Cross-Channel Color IE (CCC-IE) algorithm is presented, which has high shuffling ability with diagonal permutation and bi-directional sequential diffusion through a novel 2D hyperchaotic hybrid map of optimization test functions. The proposed 2D hyperchaotic map, namely 2D-RG map, is designed by hybridizing the Rastrigin and Griewank functions. Their chaotic nature, such as high complexity and fluctuation, makes them suitable for design of chaotic map. The 2D-RG map-based CCC-IE algorithm diagonally scrambles the pixels of the image across RGB channels and sequentially manipulates the pixel values in bi-direction for improving resistance of the algorithm to the cyberthreats. The dynamic performance of the 2D-RG map and the reliability of the CCC-IE algorithm are separately investigated and validated through a detailed comparison with the state of the art. The results manifest that the CCC-IE not only visually but also numerically is the more reliable algorithm thanks to the superior chaotic characteristics of the 2D-RG map.

Development of a deep learning model to distinguish the cause of optic disc atrophy using retinal fundus photography

The differential diagnosis for optic atrophy can be challenging and requires expensive, time-consuming ancillary testing to determine the cause. While Leber's hereditary optic neuropathy (LHON) and optic neuritis (ON) are both clinically significant causes for optic atrophy, both relatively rare in the general population, contributing to limitations in obtaining large imaging datasets. This study therefore aims to develop a deep learning (DL) model based on small datasets that could distinguish the cause of optic disc atrophy using only fundus photography. We retrospectively reviewed fundus photographs of 120 normal eyes, 30 eyes (15 patients) with genetically-confirmed LHON, and 30 eyes (26 patients) with ON. Images were split into a training dataset and a test dataset and used for model training with ResNet-18. To visualize the critical regions in retinal photographs that are highly associated with disease prediction, Gradient-Weighted Class Activation Map (Grad-CAM) was used to generate image-level attention heat maps and to enhance the interpretability of the DL system. In the 3-class classification of normal, LHON, and ON, the area under the receiver operating characteristic curve (AUROC) was 1.0 for normal, 0.988 for LHON, and 0.990 for ON, clearly differentiating each class from the others with an overall total accuracy of 0.93. Specifically, when distinguishing between normal and disease cases, the precision, recall, and F1 scores were perfect at 1.0. Furthermore, in the differentiation of LHON from other conditions, ON from others, and between LHON and ON, we consistently observed precision, recall, and F1 scores of 0.8. The model performance was maintained until only 10% of the pixel values of the image, identified as important by Grad-CAM, were preserved and the rest were masked, followed by retraining and evaluation.

On-line monitoring of membrane fouling based on an improved electrical measurement method.

On-line monitoring of membrane fouling is essential in the water treatment process. Drawbacks such as low-sensitivity and off-line limitations limit the application of existing methods. An on-line monitoring method based on Electrical Resistance Tomography (ERT) sensors is put forward in this paper. The Particle Swarm Optimization with Simulated Annealing (PSO-SA) algorithm is used in optimizing the topologies of finite element models in order to decrease the ill-posedness of sensitivity matrices. The deep denoising extreme learning machine with an auto-encoder model and the K-singular value decomposition algorithm are used in ERT reconstruction to improve imaging quality. The lift-wavelet is adopted in measuring the permeate flux to improve measuring accuracy. The ERT pixel values of the membrane module and the result of flux are used to analyze the fouling status. The results of membrane fouling experiments demonstrate the following: (1) Based on the local ERT pixels, the "two stage" phenomenon of membrane fouling can be observed. (2) In the early stage, the fouling distribution of the localized membrane module is consistent with its ERT pixels. (3) The deposition process of foulants for the localized membrane module is synchronized with the variation of ERT pixels. (4) The integrity of the membrane module can be detected according to the ERT pixels. Therefore, the novel method can effectively reflect the membrane fouling process, especially in the early stages of membrane fouling.

Swarm intention identification via dynamic distribution probability image

In the realm of decision-making for defense and security applications, it is paramount to swiftly and accurately identify the intentions of incoming swarms. Conventional identification methods predominantly focus on single-target applications and overlook the perturbations introduced by measurement noise. In this study, we propose a novel concept: the Dynamic Distribution Probability (DDP) image, which is constructed using the estimated state and its covariance matrix. Each grayscale pixel value within the image signifies the probability of the presence of the agent within the swarm. Our proposed identification scheme integrates the use of Extended Kalman Filter (EKF), Convolutional Neural Network (CNN), Back Propagation (BP) network, and Gated Recurrent Unit (GRU) network. Specifically, the DDP image is processed through a CNN to distill the formation characteristics, and the estimated swarm state from EKF is inputted into a BP network to deduce the kinematic information. The outputs from both networks are summed and subsequently channeled into a GRU network to capture temporal dynamics. Extensive numerical simulations and flight experiments are presented to demonstrate the robust anti-noise capability of the proposed scheme compared with conventional methods, as well as its superior training efficiency.

Image Transform and Hybrid Optimization Based Feature Reduction for Leaf Diseases Classification

Abstract: In recent years, there has been amazing progress in technology innovation across many sectors, including medical care, homeland security, the arts, and even agriculture. The ability to run one's business well is highly regarded in every sector. Because to recent advancements in technology, such as smart phones and the Internet of Things, this is now a distinct possibility. In this research, we offer a smart management system for horticulture that makes use of machine learning methods. This system is based on the smart management scheme that was previously described. As a component of horticulture management, many academics have advocated for the use of machine learning to analyse plant growth based on the features of the leaf. We used the grayscale image, which employed numerous features for classification while simultaneously normalising the pixel values. As a consequence of this, the processing of damaged leaves in their early stages takes more time and results in less accurate classification. As a result, for the purpose of this work, we do a colour space conversion in order to investigate the image in terms of hue, saturation, and value. Following that, it extracted features from the Value channel by making advantage of the histogram's attributes. After that, the best characteristics were chosen by a process that included both particle swarm and cuckoo search optimisation. Following this, a regression neural network was trained to classify data based on the features that had been selected in the previous step. In order to evaluate how effective the proposed strategy is, classification performance criteria such as accuracy, sensitivity, and specificity were utilised. The suggested approach was tested in a simulated Windows 10 environment using a real-time dataset and the MATLAB software. It was compared to other methods that are already in use. While evaluating performance, comparisons to the current state of the art are conducted with regard to accuracy, sensitivity, and specificity

Ultrasound-based assessment of the expression of inflammatory markers in the rectus femoris muscle of rats.

Ultrasonographic characteristics of skeletal muscles are related to their health status and functional capacity, but they still provide limited information on muscle composition during the inflammatory process. It has been demonstrated that an alteration in muscle composition or structure can have disparate effects on different ranges of ultrasonogram pixel intensities. Therefore, monitoring specific clusters or bands of pixel intensity values could help detect echotextural changes in skeletal muscles associated with neurogenic inflammation. Here we compare two methods of ultrasonographic image analysis, namely, the echointensity (EI) segmentation approach (EI banding method) and detection of selective pixel intensity ranges correlated with the expression of inflammatory regulators using an in-house developed computer algorithm (r-Algo). This study utilized an experimental model of neurogenic inflammation in segmentally linked myotomes (i.e., rectus femoris (RF) muscle) of rats subjected to lumbar facet injury. Our results show that there were no significant differences in RF echotextural variables for different EI bands (with 50- or 25-pixel intervals) between surgery and sham-operated rats, and no significant correlations among individual EI band pixel characteristics and protein expression of inflammatory regulators studied. However, mean numerical pixel values for the pixel intensity ranges identified with the proprietary r-Algo computer program correlated with protein expression of ERK1/2 and substance P (both 86-101-pixel ranges) and CaMKII (86-103-pixel range) in RF, and were greater (p < 0.05) in surgery rats compared with their sham-operated counterparts. Our findings indicate that computer-aided identification of specific pixel intensity ranges was critical for ultrasonographic detection of changes in the expression of inflammatory mediators in neurosegmentally-linked skeletal muscles of rats after facet injury.

Blind watermarking algorithm with 2D-SCCM chaotic system encryption based on SURF and RHFMs

In this paper, a digital watermarking algorithm was proposed based on local feature regions and radial harmonic Fourier moments. It used a quantized modulation algorithm to embed the watermark information into the frequency domain information of the image. Compared with traditional watermarks directly embedded into the pixel values, the algorithm enhanced the invisibility of the watermark. Many existing digital watermarking algorithms neglected the security of the embedding process, because of more attention on the effect of post-embedding and extraction. To overcome this problem, we proposed a two-dimensional hyperchaotic system named 2D-SCCM chaotic system, which used classical sine mapping and cubic mapping for coupling. Before embedding, the correlation among the pixels of the watermarked image was disrupted, resulting in a full process security enhancement. Combining the clustering algorithm with feature point extraction, the obtained local feature region was more stable and suitable for watermark embedding. Results showed that the improved watermark embedding algorithm was more effective for resisting geometric attacks such as cropping attacks. Moreover, the PSNR values of the final watermark-embedded images were all greater than 50 dB, indicating a good balance between imperceptibility and robustness.

Analyzing land cover change dynamism through a GIS-based method: application to Gran Canaria (Canary Islands, Spain)

Landscapes are dynamic areas which can be studied from the perspective of different, but related, disciplines. The application of GIS methods to study landscape change is an interesting resource to help understand and explain landscape dynamics. Thus, the objective of this paper is to characterize land cover dynamism and identify land cover change ‘hotspots’. The method used here generates raster maps in which each pixel value represents the number of times each pixel has changed land cover between 1990-2018. Furthermore, it provides three of these maps, since it works with the 3 taxonomic levels of Corine Land Cover datasets, giving us different levels of detail in the analysis. On the other hand, the statistical treatment of the data has been done at the municipal level. The most important results reveal that Agüimes is the municipality with the most dynamic land cover change in the three levels, while Tejeda, in level 1, and Valsequillo, in levels 2 and 3, have the least changing land cover. These outputs are complemented with other statistical analyses which allow the integration of different data types such as those related to population, tourism and agriculture. Subsequently, some of the methodological issues and findings are discussed and put in context with the scientific literature.

A challenge and solution for automatic thin slice thickness measurements on images of the Catphan phantom

Purpose. The use of the Hough transform for angle detection is quite accurate for relatively wide slice thickness. However, the Hough transform fails to accurately detect the angle for thin slice thickness. This study proposes a method for automatically measuring the thickness of thin slices on images of a Catphan phantom. Methods. In the proposed method, the angle of the phantom’s orientation was determined based on the relative coordinates of the four hole objects in the phantom. After the angles of the wires were determined, the profiles of pixel values across the wire objects were constructed. Finally, their full widths at half maximum (FWHMs) were determined and multiplied by tan 23° to obtain the slice thicknesses of the images. The results of the proposed method were compared to a previous method, which used the Hough transform to obtain the phantom’s orientation. We used slice thicknesses ranging from 0.8 mm to 5.0 mm, and phantom angles from 0° to 10°. Results. Our proposed method detected the angle of the phantom accurately for thin slices, whereas a previous method did not accurately detect the angle. The results of the slice thickness using this current method were slightly higher (within 7.9%) compared to the previous method. However, the results of the two methods did not differ significantly (p-value > 0.05). Using different angles, the current method detected all the angles more accurately. Again, the slice thicknesses were not significantly different from the previous method (p-value > 0.05). Conclusion. The proposed method for measuring the thickness of thin slices in an image of a Catphan phantom, based on the relative coordinates of the four hole objects in the phantom, outperformed a previous method based on the Hough transform.

Celestial Object Detection in Astronomical Images Using MSE and Jacobi Moments

Astronomical images share a common characteristic, which is low variability. This term refers to images with minimal changes or variations in pixel values or features. In this article, we leverage the low variability characteristic of astronomical images to detect celestial objects. We compute the mean squared error for the reconstructed astronomical image, derived from its Jacobi moments. These moments are calculated multiple times, focusing on different regions within the image. The mean squared error metric serves as the score function for the simulated annealing algorithm, aiding in the identification of regions with the highest information loss. The parameters used to compute Jacobi moments to focus on that region are then interpreted as the coordinates of celestial objects. This method proves effective for preprocessing images since it provides optimal parameters for Jacobi polynomials, which will enhance their feature extraction capability. Additionally, it serves as an object detection method, as we can interpret the Jacobi moments' parameters as coordinates for the objects within the images.

Face De-Identification Using Convolutional Neural Network (CNN) Models for Visual-Copy Detection

The proliferation of media-sharing platforms has led to issues with illegally edited content and the distribution of pornography. To protect personal information, de-identification technologies are being developed to prevent facial identification. Existing de-identification methods directly alter the pixel values in the face region, leading to reduced feature representation and identification accuracy. This study aims to develop a method that minimizes the possibility of personal identification while effectively preserving important features for image- and video-copy-detection tasks, proposing a new deep-learning-based de-identification approach that surpasses traditional pixel-based alteration methods. We introduce two de-identification models using different approaches: one emphasizing the contours of the original face through feature inversion and the other generating a blurred version of the face using D2GAN (Dual Discriminator Generative Adversarial Network). Both models were evaluated on their performance in image- and video-copy-detection tasks before and after de-identification, demonstrating effective feature preservation. This research presents new possibilities for personal-information protection and digital-content security, contributing to digital-rights management and law enforcement.

A fully automatic deep learning-based method for segmenting regions of interest and predicting renal function in pediatric dynamic renal scintigraphy.

Accurate delineation of renal regions of interest (ROIs) is critical for the assessment of renal function in pediatric dynamic renal scintigraphy (DRS). The purpose of this study was to develop and evaluate a deep learning (DL) model that can fully automatically delineate renal ROIs and calculate renal function in pediatric 99mTechnetium-ethylenedicysteine (99mTc-EC) DRS. This study retrospectively analyzed 1,283 pediatric DRS data at a single center from January to December 2018. These patients were divided into training set (n = 1027), validation set (n = 128), and testing set (n = 128). A fully automatic segmentation of ROIs (FASR) model was developed and evaluated. The pixel values of the automatically segmented ROIs were calculated to predict renal blood perfusion rate (BPR) and differential renal function (DRF). Precision, recall rate, intersection over union (IOU), and Dice similarity coefficient (DSC) were used to evaluate the performance of FASR model. Intraclass correlation (ICC) and Pearson correlation analysis were used to compare the consistency of automatic and manual method in assessing the renal function parameters in the testing set. The FASR model achieved a precision of 0.88, recall rate of 0.94, IOU of 0.83, and DSC of 0.91. In the testing set, the r values of BPR and DRF calculated by the two methods were 0.94 (P < 0.01) and 0.97 (P < 0.01), and the ICCs (95% confidence interval CI) were 0.94 (0.90-0.96) and 0.94 (0.91-0.96). We propose a reliable and stable DL model that can fully automatically segment ROIs and accurately predict renal function in pediatric 99mTc-EC DRS.

Image Steganography and Style Transformation Based on Generative Adversarial Network

Traditional image steganography conceals secret messages in unprocessed natural images by modifying the pixel value, causing the obtained stego to be different from the original image in terms of the statistical distribution; thereby, it can be detected by a well-trained classifier for steganalysis. To ensure the steganography is imperceptible and in line with the trend of art images produced by Artificial-Intelligence-Generated Content (AIGC) becoming popular on social networks, this paper proposes to embed hidden information throughout the process of the generation of an art-style image by designing an image-style-transformation neural network with a steganography function. The proposed scheme takes a content image, an art-style image, and messages to be embedded as inputs, processing them with an encoder–decoder model, and finally, generates a styled image containing the secret messages at the same time. An adversarial training technique was applied to enhance the imperceptibility of the generated art-style stego image from plain-style-transferred images. The lack of the original cover image makes it difficult for the opponent learning steganalyzer to identify the stego. The proposed approach can successfully withstand existing steganalysis techniques and attain the embedding capacity of three bits per pixel for a color image, according to the experimental results.

Differentiating the Presence of Brain Stroke Types in MR Images using CNN Architecture.

Stroke is reportedly the biggest cause of death and disability in the world, according to the World Health Organisation (WHO). The severity of a stroke can be lowered by recognising the many stroke warning indicators early on. Using CNN, the primary goal of this study is to predict the likelihood that a brain stroke would develop at an early stage. The novelty of the proposed work is to acquire models that can accurately differentiate between stroke and no-stroke (normal) cases using MR Imaging sequences like DWI, SWI, GRE and T2 FLAIR aiding in timely diagnosis and treatment decisions. A dataset comprising real time MRI scans of patients with stroke and no-stroke conditions was collected and preprocessed for model training. The preprocessing involves standardizing the resolution of the images, normalizing pixel values, and augmenting the dataset to enhance model generalization. The ResNet, DenseNet, EfficientNet, and VGG16 architectures were implemented and trained on the preprocessed dataset. The training process involved optimizing model parameters using stochastic gradient descent and minimizing the loss function. The results demonstrate promising performance across all models by obtaining an accuracy of 98% for ResNet, DenseNet and EfficientNet, while 97% for VGG16 in differentiating the stroke using real time MRI data. The proposed work explored the effectiveness of CNN models, including ResNet, DenseNet, EfficientNet, and VGG16, for the differentiation of stroke and no-stroke cases. The models were trained and evaluated using a real-time dataset of brain MR Images. The obtained accuracies highlight the potential of CNN models in accurately differentiating between stroke and non-stroke cases.

Deep Learning-Based Detection of Defects in Wafer Buffer Zone During Semiconductor Packaging Process

Effectively detecting defects in wafer buffer zones is crucial in semiconductor manufacturing processes. If defects are not detected in the middle of the semiconductor manufacturing process, the semiconductor eventually becomes a defective product after all processes are completed. To solve this problem, rule-based vision algorithms have been used to identify defects in the wafer buffer zone. After photographing the wafer buffer zone using a high-speed camera, defects are detected using the pixel values of the images. However, because of the resin bleed in the wafer, which is an epoxy compound, it is difficult to detect defects. Therefore, we introduced a deep learning method for semiconductor inspection and created a high-performance semiconductor inspection algorithm. The defects in the wafer buffer zone should be detected accurately and quickly. Furthermore, the approximate size of the defects must be extracted. We modified the Xception model to fit the wafer data characteristics considering both accuracy and speed. We proposed to extract the size of the defect using class activation mapping (CAM). We obtained an accuracy of 96.9% from the actual wafer dataset through this framework, and then managed to extract the size of the defect through CAM.