Search

Abstract We propose a novel framework that combines state-of-the-art deep learning approaches with pre- and post-processing algorithms for particle detection in complex/heterogeneous backgrounds common in the manufacturing domain. Traditional methods, like size analyzers and those based on dilution, image processing, or deep learning, typically excel with homogeneous backgrounds. Yet, they often fall short in accurately detecting particles against the intricate and varied backgrounds characteristic of heterogeneous particle–substrate (HPS) interfaces in manufacturing. To address this, we've developed a flexible framework designed to detect particles in diverse environments and input types. Our modular framework hinges on model selection and AI-guided particle detection as its core, with preprocessing and postprocessing as integral components, creating a four-step process. This system is versatile, allowing for various preprocessing, AI model selections, and post-processing strategies. We demonstrate this with an entrainment-based particle delivery method, transferring various particles onto substrates that mimic the HPS interface. By altering particle and substrate properties (e.g., material type, size, roughness, shape) and process parameters (e.g., capillary number) during particle entrainment, we capture images under different ambient lighting conditions, introducing a range of HPS background complexities. In the preprocessing phase, we apply image enhancement and sharpening techniques to improve detection accuracy. Specifically, image enhancement adjusts the dynamic range and histogram, while sharpening increases contrast by combining the high pass filter output with the base image. We introduce an image classifier model (based on the type of heterogeneity), employing Transfer Learning with MobileNet as a Model Selector, to identify the most appropriate AI model (i.e., YOLO model) for analyzing each specific image, thereby enhancing detection accuracy across particle–substrate variations. Following image classification based on heterogeneity, the relevant YOLO model is employed for particle identification, with a distinct YOLO model generated for each heterogeneity type, improving overall classification performance. In the post-processing phase, domain knowledge is used to minimize false positives. Our analysis indicates that the AI-guided framework maintains consistent precision and recall across various HPS conditions, with the harmonic mean of these metrics comparable to those of individual AI model outcomes. This tool shows potential for advancing in-situ process monitoring across multiple manufacturing operations, including high-density powder-based 3D printing, powder metallurgy, extreme environment coatings, particle categorization, and semiconductor manufacturing.

In the realm of digital commerce, online shopping has witnessed unprecedented growth globally, becoming a cornerstone of modern consumer behavior. This research introduces an advanced customer segmentation model, named LRFS, which builds upon the traditional LRF framework (Length of Relationship, Recency of Purchase, and Frequency of Purchase), specifically tailored for the e-commerce sector. The innovation of the LRFS model lies in the integration of a novel component, “S”, which quantifies the Staying Rate relative to the revenue generated by customers on a specific website. This addition aims to enhance the granularity and efficacy of customer segmentation by leveraging data extracted from Google Analytics. To operationalize the LRFS model, this study employs two renowned clustering algorithms, K-Means and K-Medoids, analyzing the dataset through the lens of three distinct dimensionality reduction techniques: PCA (Principal Component Analysis), t-SNE (t-Distributed Stochastic Neighbor Embedding), and Autoencoder. This methodological approach facilitates a robust comparative analysis between the LRFS model and its predecessors — LR, LF, and LRF — utilizing K-Means clustering to evaluate the precision of customer cluster assignments. The empirical findings of this research underscore the superiority of the LRFS model in achieving more accurate and insightful customer segmentation. Additionally, a composite Customer Classification and Customer Relationship Matrix was deployed to discern the nuanced traits of clustered groups, identifying the fusion of K-Medoids and t-SNE as the most effective strategy for capturing the full spectrum of customer dynamics. The research further elucidates several test cases and use case scenarios, demonstrating the practical applications of the LRFS model in conjunction with K-Means and PCA to refine marketing strategies and foster a deeper understanding of the online customer base. Through the development and application of the LRFS model, this study contributes significantly to the field of e-commerce by providing a more nuanced tool for businesses to tailor their marketing initiatives, ensuring alignment with the evolving preferences and behaviors of their online clientele.

A physical cyber system connects all authenticated cyber devices in its network. Nowadays, many wearable devices function as cyber devices. In essence, people are using these devices more for their healthcare. These devices would be very popular if an easy-to-use manufacturing architecture could be created and, at the same time, the devices could protect the data stored on the device. In this article, we suggest a good manufacturing architecture for a healthcare device, in which our proposed data protection method works well. The architecture is very simple to implement and the data protection method hides information in a DNA sequence. The present DNA-based data hiding schemes implant secrets in converted binaries of nucleotides. The number of implanted bits is no more than the length of nucleotides, however, these schemes expand the stego DNA sequence noticeably. While implanting a large message, e.g., historical records of patients, it would be harder for these schemes to manage the implantation of the whole secret in a single DNA sequence. A large DNA sequence might be a solution in some contexts. Nevertheless, managing a large DNA sequence and its expanded part in a fixed memory space would be challenging as those are too large in size. To address this problem, we propose a multi-time data embedment method that could implant as much data in a DNA sequence as needed. Although it presents a greater embedding capacity, it does not increase the length of the modified DNA/RNA sequence. Thus, it optimizes the extra memory load in the chip. The proposed method implemented several features to improve the security of both implanted data and DNA sequence. The experimental results outperform all measurements over the competing schemes.

Brain cancer classification is an important step that depends on the physician's knowledge and experience. An automated tumor classification system is very essential to support radiologists and physicians to identify brain tumors. However, the accuracy of current systems needs to be improved for suitable treatments. In this paper, we propose a hybrid feature extraction method with a regularized extreme learning machine (RELM) for developing an accurate brain tumor classification approach. The approach starts by preprocessing the brain images by using a min-max normalization rule to enhance the contrast of brain edges and regions. Then, the brain tumor features are extracted based on a hybrid method of feature extraction. Finally, a RELM is used for classifying the type of brain tumor. To evaluate and compare the proposed approach, a set of experiments is conducted on a new public dataset of brain images. The experimental results proved that the approach is more effective compared with the existing state-of-the-art approaches, and the performance in terms of classification accuracy improved from 91.51% to 94.233% for the experiment of the random holdout technique.