Traditional Image Processing Research Articles

Deep transfer learning for fine-grained maize leaf disease classification

Machine learning (ML) can enhance agricultural yields by combating plant diseases and climate change. However, traditional image processing techniques for disease detection have limitations in robustness and generalizability. In this study, we investigate deep transfer learning for fine-grained disease classification in maize plants, which is a challenging task due to the subtle and nuanced disease patterns. We use four tailored deep learning frameworks: VGGNET, Inception V3, ResNet50, and InceptionResNetV2. ResNet50 achieves the highest validation accuracy of 87.51%, precision of 90.33%, and recall of 99.80%, demonstrating the efficacy and superiority of our approach. Our study offers an innovative solution for accurate disease classification in maize plants.

Crack _ PSTU: Crack detection based on the U-Net framework combined with Swin Transformer

Cracks serve as a significant indicator of aging infrastructure. While the surfaces of traditional infrastructures are visually inspected manually, this approach is labor-intensive, highly subjective, and often surpasses the capabilities of available inspection personnel. Although some researchers have employed traditional image processing and machine learning techniques to address these challenges, the presence of irregular crack shapes, complex lighting conditions, and limitations in detecting only a single type of surface crack complicate automatic crack detection. Recent research indicates that deep learning methods are increasingly leading in image-based feature extraction, object detection, and pixel attribute analysis. This paper introduces a method, Crack_PSTU (Pre-trained Swin Transformer U-Net), which leverages the U-Net framework and Swin Transformer model for infrastructure crack classification and detection tasks. Specifically, we utilize a pre-trained Swin Transformer network as the encoder and implement the decoder using convolution, pooling, and other operations, creating a "U"-shaped model architecture. Our experiments encompassed 11,298 crack images from varied scenes, with 9603 designated for training and 1695 for validation. The findings reveal that our approach surpasses other algorithms for this dataset.

A survey on brain MRI segmentation

Neurological disorders pose a significant challenge in medical diagnostics, requiring accurate and efficient analysis of brain Magnetic Resonance Imaging (MRI) scans. This study introduces a novel approach for enhanced neurological diagnosis through the application of deep learning techniques to automate the segmentation of brain structures in MRI images. The proposed method leverages the power of convolutional neural networks (CNNs) to extract intricate patterns and features from complex neuro imaging data. The research involves the development and training of a deep learning model capable of accurately delineating key anatomical regions, such as the cortex, hippocampus, and ventricles. The model is trained on a large data set of annotated MRI scans, optimizing its performance through rigorous validation processes. The utilization of deep learning enables the algorithm to learn and generalize from diverse imaging data, improving its adaptability to variations in patient demographics and scanner characteristics. To evaluate the effectiveness of the proposed approach, comprehensive experiments are conducted on a diverse set of MRI datasets, encompassing various neurological conditions. Quantitative metrics, including Dice coefficient and Hausdorff distance, are employed to assess the segmentation accuracy compared to ground truth annotations. Additionally, the clinical relevance of the automated segmentation is validated through collaboration with neurologists and radiologists. The results demonstrate that the deep learning- enabled segmentation method consistently outperforms traditional image processing techniques, providing more accurate and reliable segmentation results. The proposed approach not only streamlines the diagnostic process but also has the potential to uncover subtle abnormalities that may be overlooked by manual inspection. In conclusion, the integration of deep learning into the segmentation of brain MRI scans presents a promising avenue for enhancing neurological diagnosis. The automated and precise delineation of brain structures contributes to the efficiency and accuracy of diagnostic workflows, ultimately improving patient outcomes and facilitating timely interventions in the realm of neurological disorders.

Research on image processing of electric power system terminals based on reinforcement learning and mobile edge computing optimization

AbstractThis research is dedicated to the optimization of power system terminal image processing based on RL and MEC. With the continuous development of power system, the demand for image processing of terminal equipment is increasing day by day. However, traditional image processing methods have the problems of high computing complexity and real‐time and energy consumption. To solve this problem, this study introduces the idea of RL and MEC to improve the efficiency and performance of image processing of power system terminals. By modeling and optimizing the image processing task of the power system terminal equipment, the intelligent adjustment of the processing parameters is realized to adapt to the needs of different scenarios. MEC technology is introduced to move image processing tasks from the central server to the edge device, reducing data transmission delay and network burden, thus improving real‐time performance and reducing energy consumption. The experimental results show that the proposed optimization method based on RL and MEC has a significant performance improvement compared with the traditional method in the power system terminal image processing. The framework our proposed has achieved significant improvement in task completion latency, achieving higher system energy efficiency compared to traditional methods.

Image-based automatic multiple-defect detection of urban utility tunnel using UUTNet

In-service urban utility tunnels (UUT) suffer from cracks, corrosion, and leakage defects, which rises the chance of major accidents. However, prevailing detection methods for UUT remain reliant on manual inspection and subjective judgment, or traditional image processing technologies, such methods may not be able to obtain accurate defect information. This study proposes a novel and effective network called UUTNet based on the constructed UUT dataset for defects detection. Considering that the UUT defects has a certain distribution correlation, the attention module is introduced to the Pyramid Scene Parsing Network to capture the relation. By adding the hybrid dilated convolution after the feature extraction layer, the receptive field is expanded to further extract global and local features. The performance of UUTNet was evaluated based on the metrics MIoU, F1-score, Accuracy, and robustness. Comparative experiments were conducted, and the results showed the UUTNet achieved the best detection performance, achieving 0.7615 MIoU, 0.9806 Accuracy and 0.8012 F1-score. The MIoU was further improved to 0.7847 by utilizing the Bayesian optimization. Three extreme inspection scenes, including uneven illumination, high brightness, and obstacle interference, were applied to validate model robustness. The proposed method offers robust technical assistance for detecting defects in the UUT and precisely assessing the distribution and extent of these defects.

QR code resolution improvement based on Super-Resolution Generative Adversarial Network

QR codes have become an integral part of our daily routines, simplifying tasks ranging from accessing websites to making payments. However, the quality of QR codes, especially their resolution, can significantly impact their functionality. Low-resolution QR codes may lead to misinterpretation during scanning and even decoding failures. To address this issue, researchers have explored various techniques to enhance the resolution of QR codes. Traditional image processing methods, such as interpolation and filtering, have been used in the past for resolution enhancement. However, these methods often result in overly blurry images with poor perceptual quality. Conversely, solutions based on Convolutional Neural Networks (CNNs) can introduce clarity but may compromise the sharpness of image edges. This paper presents an effective approach to improve QR code resolution using a Super-Resolution Generative Adversarial Network (SRGAN). The results are impressive, with SRGAN achieving a Peak Signal-to-Noise Ratio (PSNR) of 30.06, significantly outperforming the 17.48 achieved by the SRCNN method. Additionally, in terms of Structural Similarity Index (SSIM), SRGAN reaches 0.936, surpassing SRCNN's 0.473. These metrics demonstrate that SRGAN is highly effective in enhancing the resolution of QR codes, ensuring better scan accuracy and overall functionality in practical applications.

EyeMo: A Solution for Individuals with Disabilities to Use a Computer Through Eye Movements

The maintenance of an upright posture and the execution of desired movements by individuals necessitate the seamless and harmonious functioning of the muscular and skeletal systems. Neuromuscular diseases, spinal cord injuries, and related conditions can adversely impact individuals' motor functions, leading to a loss of muscle strength and, in severe cases, paralysis. Such health challenges may result in social isolation and detachment from one's social life. This study is focused on the development of a computer control application utilizing eye movements to aid users in navigating and interacting with computers. The system utilizes a lightweight eyeglass frame and a mini-USB camera to accurately capture images of the user's pupil. Pupil detection is achieved through a combination of traditional image processing and deep learning techniques, ensuring high accuracy under diverse conditions. Users have the ability to interactively calibrate the application to accommodate varying screen sizes, thereby enhancing system responsiveness. The user interface incorporates functionalities such as left-click, right-click, double-click, and scrolling, empowering users to perform tasks ranging from internet browsing to video watching. 
 After the development of the application, a request for research permission was submitted to the local institution to allow volunteers to test the application, adhering to ethical standards. Following the approval of ethical clearance, the application was made available for use by volunteers, and feedback was systematically collected. Volunteers generally reported that the application was beneficial, indicating its potential use by disadvantaged individuals. The upcoming versions of the application have the potential for integration with smart home systems. Additionally, the developed application can be used in games and for educational purposes.

Scaled aggregation operations over three-dimensional extended intuitionistic fuzzy index matrices

Modern image processing techniques are improving beyond old methods, which include advanced approaches, for example deep learning. Convolutional Neural Networks (CNNs) are excellent at automatic feature extraction, whereas Generative Adversarial Networks (GANs) produce realistic images. Transfer learning uses pre-trained models, whereas semantic segmentation identifies pixels in images. Super-resolution, style transfer, and attention mechanisms can increase the quality of images and understanding. Adversarial defenses address purposeful manipulations, while 3D image processing handles three-dimensional data. These advancements make use of improved computational power and massive datasets to revolutionize image processing capabilities. Traditional image processing algorithms frequently fail to handle the complex and multidimensional structure of color images, particularly when dealing with uncertainty and imprecision. In this study, the 3D-EIFIM frame work is extented and scaled aggregation operations 3D-EIFIM tailored for image data are proposed. By representing each pixel as an entry of 3D-EIFIM and applying aggregation techniques to enable more effective image analysis, manipulation, and enhancement. The practical implications of this research are significant, as it can lead to advancements in fields such as computer vision, medical imaging, and remote sensing.

Small object detection algorithm incorporating swin transformer for tea buds.

Accurate identification of small tea buds is a key technology for tea harvesting robots, which directly affects tea quality and yield. However, due to the complexity of the tea plantation environment and the diversity of tea buds, accurate identification remains an enormous challenge. Current methods based on traditional image processing and machine learning fail to effectively extract subtle features and morphology of small tea buds, resulting in low accuracy and robustness. To achieve accurate identification, this paper proposes a small object detection algorithm called STF-YOLO (Small Target Detection with Swin Transformer and Focused YOLO), which integrates the Swin Transformer module and the YOLOv8 network to improve the detection ability of small objects. The Swin Transformer module extracts visual features based on a self-attention mechanism, which captures global and local context information of small objects to enhance feature representation. The YOLOv8 network is an object detector based on deep convolutional neural networks, offering high speed and precision. Based on the YOLOv8 network, modules including Focus and Depthwise Convolution are introduced to reduce computation and parameters, increase receptive field and feature channels, and improve feature fusion and transmission. Additionally, the Wise Intersection over Union loss is utilized to optimize the network. Experiments conducted on a self-created dataset of tea buds demonstrate that the STF-YOLO model achieves outstanding results, with an accuracy of 91.5% and a mean Average Precision of 89.4%. These results are significantly better than other detectors. Results show that, compared to mainstream algorithms (YOLOv8, YOLOv7, YOLOv5, and YOLOx), the model improves accuracy and F1 score by 5-20.22 percentage points and 0.03-0.13, respectively, proving its effectiveness in enhancing small object detection performance. This research provides technical means for the accurate identification of small tea buds in complex environments and offers insights into small object detection. Future research can further optimize model structures and parameters for more scenarios and tasks, as well as explore data augmentation and model fusion methods to improve generalization ability and robustness.

Image Processing Techniques for Improving Quality of 3D Profile in Digital Holographic Microscopy Using Deep Learning Algorithm.

Digital Holographic Microscopy (DHM) is a 3D imaging technology widely applied in biology, microelectronics, and medical research. However, the noise generated during the 3D imaging process can affect the accuracy of medical diagnoses. To solve this problem, we proposed several frequency domain filtering algorithms. However, the filtering algorithms we proposed have a limitation in that they can only be applied when the distance between the direct current (DC) spectrum and sidebands are sufficiently far. To address these limitations, among the proposed filtering algorithms, the HiVA algorithm and deep learning algorithm, which effectively filter by distinguishing between noise and detailed information of the object, are used to enable filtering regardless of the distance between the DC spectrum and sidebands. In this paper, a combination of deep learning technology and traditional image processing methods is proposed, aiming to reduce noise in 3D profile imaging using the Improved Denoising Diffusion Probabilistic Models (IDDPM) algorithm.

Algorithms for Liver Segmentation in Computed Tomography Scans: A Historical Perspective.

Oncology has emerged as a crucial field of study in the domain of medicine. Computed tomography has gained widespread adoption as a radiological modality for the identification and characterisation of pathologies, particularly in oncology, enabling precise identification of affected organs and tissues. However, achieving accurate liver segmentation in computed tomography scans remains a challenge due to the presence of artefacts and the varying densities of soft tissues and adjacent organs. This paper compares artificial intelligence algorithms and traditional medical image processing techniques to assist radiologists in liver segmentation in computed tomography scans and evaluates their accuracy and efficiency. Despite notable progress in the field, the limited availability of public datasets remains a significant barrier to broad participation in research studies and replication of methodologies. Future directions should focus on increasing the accessibility of public datasets, establishing standardised evaluation metrics, and advancing the development of three-dimensional segmentation techniques. In addition, maintaining a collaborative relationship between technological advances and medical expertise is essential to ensure that these innovations not only achieve technical accuracy, but also remain aligned with clinical needs and realities. This synergy ensures their applicability and effectiveness in real-world healthcare environments.

Applied Deep learning approaches on canker effected leaves to enhance the detection of the disease using Image Embedding and Machine learning Techniques

Canker, a disease that causes considerable financial losses in the agricultural business, is a small deep lesion that is visible on the leaves of many plants, especially citrus/apple trees. Canker detection is critical for limiting its spread and minimizing harm. To address this issue, we describe a computer vision-based technique that detects Canker in citrus leaves using image embedding and machine learning (ML) algorithms. The major steps in our proposed model include image embedding, and machine learning model training and testing. We started with preprocessing and then used image embedding techniques like Inception V3 and VGG 16 to turn the ROIs into feature vectors that retained the relevant information associated with Canker leaf disease, using the feature vectors acquired from the embedding stage, we then train and evaluate various ML models such as support vector machines (SVM), Gradient Boosting, neural network, and K Nearest Neighbor. Our experimental results utilizing a citrus leaf picture dataset show that the proposed strategy works. With Inception V3 as the image embedder and neural network machine learning model we have obtained an accuracy of 95.6% which suggests that our approach is effective in canker identification. Our method skips traditional image processing techniques that rely on by hand features and produces results equivalent to cutting-edge methods that use deep learning models. Finally, our proposed method provides a dependable and efficient method for detecting Canker in leaves. Farmers and agricultural specialists can benefit greatly from early illness diagnosis and quick intervention to avoid disease spread as adoption of such methods can significantly reduce the losses incurred by farmers and improve the quality of agricultural produce.

Underwater image enhancement by using amalgamation of colour correction, contrast-enhancing and dehazing

Underwater images can be captured either with the help of light waves or sound waves. Images that are taken underwater typically are not of optimum quality as they suffer from issues such as low contrast, blurring of detail, colour distortion, and greenish tones. Several physical processes that take place in the aquatic environment, such as light absorption, refraction, and scattering, are responsible for the existence of such degradation in underwater images. To address these challenges, numerous researchers have put forth a range of cutting-edge techniques for enhancing and restoring such degraded underwater images, with the aim of addressing these issues. These techniques primarily focus on improving visibility and enhancing the level of detail. To achieve this, we propose a method that performs White Balancing in the LAB colour space to remove the bluish-greenish tones present in the image. Next, we enhance the contrast by first converting the RGB image into HSV and HLS colour spaces and then by using the S & V channels in HSV and L & S colour channels in HLS, we apply Contrast Limited Adaptive Histogram Equalization (CLAHE). To control the brightness of the enhanced image, we apply Gamma Correction. Lastly, by using the method Dark Channel Prior (DCP), we separate the image’s red channel from the RGB colour space and perform the dehazing operation to get the final enhanced image. We have conducted a comprehensive qualitative analysis of our proposed approach as well as existing techniques, evaluating them objectively and subjectively through metrics such as peak signal-to-noise ratio (PSNR), root-mean-square error (RMSE), structural similarity (SSIM), and the underwater colour image quality evaluation metric (UCIQE) and underwater image quality measure (UIQM). Since our proposed approach uses traditional image processing methods, it is computationally less expensive and quicker as compared to deep learning or frequency domain-based methods. With this, it can be adapted for using in real-time applications such as underwater navigation, examination of the behavior of marine ecosystems and other scientific research.

Tunable optical spatial differential operation via photonic spin Hall effect in a Weyl semimetal.

Optical differential operation is the basic principle of optical image edge detection, which has the advantages of high efficiency, simple structure and markerless compared with the traditional digital image processing methods. In this paper, we propose an optical differential operation with high contrast based on the photonic spin Hall effect in a Weyl semimetal, which enables to switch between one- and two-dimensional edge detection. Due to the unique optical and electrical properties of the Weyl semimetal, a transport model for the differential operation is established, which is closely related to the beam shifts. By tuning the incidence conditions, we effectively manipulate the in-plane and transverse shifts to switch differential operations between one and two dimensions. The contrast of the differential operation is further regulated by changing the physical parameters of the Weyl semimetal, and can be improved by two orders of magnitude compared to the conventional differentiator. This study provides new possibilities in edge detection and image processing owing to the advantages of switchable dimension and high contrast.

AN EFFICIENT IMAGE PROCESSING BASED IMAGE TO CARTOON GENERATION BASED ON DEEP LEARNING

This paper proposes an approach to convert real life images into cartoon images using image processing. The cartoon images have sharp edges, reduced colour quantity compared to the original image, and smooth colour regions. With the rapid advancement in artificial intelligence, recently deep learning methods have been developed for image to cartoon generation. Most of these methods perform extremely huge computations and require large datasets and are time consuming, unlike traditional image processing which involves direct manipulation on the input images. In this paper, we have developed an image processing based method for image to cartoon generation. Here, we perform parallel operations of enhancing the edges and quantizing the colour. The edges are extracted and dilated to highlight them in the output colour image. For colour quantization, the colours are assigned based on proposed formulation on separate colour channels. Later, these images are combined and the highlighted edges are added to generate the cartoon image. The generated images are compared with existing image processing approaches and deep learning based methods. From the experimental results, it is evident that the proposed approach generates high quality cartoon images which are visually appealing, have superior contrast and are able to preserve the contextual information at lower computational cost.

A Survey on Efficient Analysis of Soft Tissues Tumors using Machine Learning

Soft tissue tumors (STT) represent a significant medical challenge, requiring accurate and efficient analysis for timely diagnosis and treatment. This survey explores the application of machine learning techniques in the analysis of soft tissue tumors, focusing on enhancing efficiency in detection and classification. The review encompasses various approaches, including traditional image processing methods and the more recent advancements in machine learning. One of the key contributions of this survey is the proposal of a method employing Convolutional Neural Networks (CNN) for the analysis of soft tissue tumors. CNNs have demonstrated remarkable success in image-related tasks, making them particularly suitable for medical image analysis. The research particularly aims in distinguishing Melanoma from normal skin tissue. To enhance the efficiency of research on the analysis of soft tissue tumors, the proposed study provides a comprehensive review of relevant literature, focusing on the application of machine learning in the identification of soft tissue tumors. This review includes an evaluation of the merits and demerits of each system. Furthermore, the study introduces a suggested model capable of providing accurate classification for soft tissue tumors.

A method for detecting spatiotemporal patterns of cancer biomarkers-evoked activity using radial basis function network extracted time-domain features from calcium imaging data

BackgroundTwo-photon calcium imaging is widely used to study the odor-evoked glomerular activity in the dorsal olfactory bulb of macrosmatic animals. The nonstationary character of activated patterns sets a limit on the use of a traditional image processing approaches. New methodThe developed method makes it possible to automatically map cancer biomarkers-activated glomeruli in the rat dorsal olfactory bulb. We interpolated fluorescence intensity of calcium dynamics based on the Gaussian RBF network and synthesized the physiological fluorescence model of the receptive glomerular field. ResultsThe experiments on 5 rats confirmed the correctness of the developed approach. Patterns evoked by the 6-methyl-5-hepten-2-one (stomach cancer biomarker) and benzene (lung cancer biomarker) were correctly identified. Comparison with existing methodsThe proposed method was compared with the nonnegative matrix factorization method and with the method based on computer vision algorithms. The developed approach showed better accuracy in experiments and provided the mathematical models of the odor-evoked patterns synthesis. These models can be used to generate synthetic images of odor-evoked glomerular activity and thus to overcome the problem of small experimental data collected in calcium imaging. ConclusionsThe proposed method should be considered part of the toolkit for fully automatic analysis of calcium imaging-based studies. Currently available methodology is not able to use breath biomarkers to reliably discriminate between cancer patients and healthy controls. Nevertheless, the effective identification of the spatial patterns of cancer biomarkers-evoked glomerular activity can serve as the foundation for highly sensitive biohybrid systems for cancer screening.

Microstructural Feature Extraction by a Convolutional Neural Network for Cold Spray of Aluminum Alloys

The use of process–microstructure–property relationships for cold spray can significantly reduce application development cost and time compared to legacy trial and error strategies. However, due to the heterogeneous microstructure of a cold spray deposit, with (prior) particle boundaries outlining consolidated splats (deformed particles) in the as-spray condition, the use of automated analysis methods is challenging. In this work, we demonstrate the utility of quantitative data developed from a convolutional neural network (CNN) for feature extraction of cold spray microstructures. Specifically, the power of CNN is harnessed to automatically segment the deformed particles, which is hardly accessible at scale with traditional image processing techniques. Deposits produced with various processing conditions are evaluated with metallography. Parameters related to particle morphology such as flattening ratio are also quantified and correlated to strength.

Street view imagery: AI-based analysis method and application

Street view imagery is an emerging form of geographic big data. It presents urban visual environments from the perspective of urban residents and also contains non-visual environment of cities, such as urban human activities and socio-economic development. However, traditional digital image processing has its limitations, and the continuous development of artificial intelligence, especially computer vision and deep learning, provides strong technical support for exploring the rich semantic information in street view imagery. This paper reviews the related research on street view imagery and its artificial intelligence analysis methods and applications. It outlines the acquisition, storage, and common data sources of street view imagery. Then it introduces computer vision, deep learning, and commonly used open-source datasets in street view imagery analysis. It also detailed three aspects of AI-based street view imagery applications, namely quantification of the physical space, urban perception, and spatial semantic speculation. Finally, issues like data acquisition, domain adaption and deep learning black box are discussed. The hotspots and prospects for the development of this research topic are also prospected.

Deep learning for thyroid nodule examination: a technical review

In recent years, the incidence of thyroid cancer has increased dramatically, resulting in an increased demand for early thyroid nodule examination. Ultrasound (US) imaging is the modality most frequently used to image thyroid nodules; However, the low image resolution, speckle noise, and high variability make it difficult to utilize traditional image processing techniques. Recent advances in deep learning (DL) have increased research into the automated processing of thyroid US images. We review three main image processing tasks for thyroid nodule analysis: classification, segmentation, and detection. We discuss the advantages and limitations of the recently proposed DL techniques as well as the data availability and algorithmic efficacy. In addition, we investigate the remaining obstacles and future potential for automated analysis of thyroid US images.