This research concentrates on the diagnosis of common mango leaf diseases in Bangladesh using image processing via deep learning. Mango production could be raised by at least 28% globally if the crop could be safeguar...
详细信息
This research concentrates on the diagnosis of common mango leaf diseases in Bangladesh using image processing via deep learning. Mango production could be raised by at least 28% globally if the crop could be safeguarded from a variety of diseases. However, without the assistance of an expert, it is challenging for the farmer to detect the disease at the appropriate time. Few studies have been conducted to identify the mango leaf disease present in Bangladesh. So far, no study has been done to identify the seven distinct mango leaf diseases reported in Bangladesh. We proposed a lightweight convolutional neural network (LCNN) in this paper to accurately classify seven distinct mango leaf diseases as well as normal mango leaf. To assess the proposed LCNN model, performance is compared to several pre-trained models such as VGG16, Resnet50, Resnet101, and Xception, and it is found that LCNN achieves the highest testing accuracy (98%).
Secure computing necessitates hardware root of trust (RoT) integrated in Systems-on-Chips (SoCs) for cryptographic keys generation, authentication and identification. In this paper, we observe that bitflips in SRAM ce...
Secure computing necessitates hardware root of trust (RoT) integrated in Systems-on-Chips (SoCs) for cryptographic keys generation, authentication and identification. In this paper, we observe that bitflips in SRAM cells that appear while accessing multiple cells from the same bitline, are not stochastic, as previously considered, but systematic. Based on this observation, a novel strong in-memory Physical Unclonable Function (PUF) computation is proposed for harvesting static entropy from SRAM arrays. The proposed design is compatible with existing in-SRAM computing architectures. To verify our PUF operation, we implement a 6T SRAM array model that performs in-memory computing using a 32 nm CMOS Technology, and, through SPICE simulation, we evaluate the proposed PUF performance. The proposed PUF operation achieves uniqueness and uniformity of 49.99%, and 49.74%, respectively, and reliability higher than 97.4% when the temperature is varied from 0°C to 100°C, and higher than 95.2% when the nominal voltage supply is varied by 10%. Furthermore, we explore the scaling of the number of Challenge Response Pairs (CRPs) of the proposed PUF, and we compare it against the state-of-the-art. Our PUF offers orders of magnitude higher number of CRPs, therefore it is suitable for integrated mechanisms that assure secure computing in SoCs.
This study explores the use of Genetic Algorithms (GA) to solve the NP-hard combinatorial optimization problem known as the Travelling Salesman Problem (TSP). The suggested GA, called GA-P, performs better than conven...
详细信息
ISBN:
(数字)9798331509675
ISBN:
(纸本)9798331509682
This study explores the use of Genetic Algorithms (GA) to solve the NP-hard combinatorial optimization problem known as the Travelling Salesman Problem (TSP). The suggested GA, called GA-P, performs better than conventional techniques because it uses a new randomized crossover threshold. According to experimental assessments, GA-P outperforms a normal Genetic Algorithm with fixed crossover (GA-N, 628.6) and approaches the best Branch and Bound solutions (B&B, 560.6) with an average cost of 617.4 for 16-node graphs. With average execution times of 0.1347 seconds for 16 nodes, GA-P also demonstrates constant computational efficiency, outperforming B&B's exponentially growing temporal complexity (169.37 seconds for 16 nodes). In order to balance solution quality and computational cost, hyperparameter tweaking revealed the ideal values of population percentage (pp = 1.1), crossover proportion (cp = 0.8), mutation threshold (mt = 0.3), and exploration probability (ep = 0.2). The results show that GA-P is a scalable and efficient substitute for TSP, with room for improvement via parallelization and exploratory techniques.
For many developing nations, agriculture is the fundamental source of economic engine. Without a significant increase in food production, the rise in global population during the 21st century would not have been conce...
详细信息
Land cover classification research contributes significantly to sustainable development by enhancing the baseline for urban planning, natural resource management, and environmental monitoring at the local, regional, n...
详细信息
A constructive way to assist surgeons before performing brain tumor surgery is by visualizing a three-dimensional (3D) MRI image to determine the brain tumor volume when a pre-operative examination. However, the avail...
详细信息
The limited autonomy of flight has long been considered a significant constraint in drone systems. In the context of drone inspections of power lines, this study focuses on a drone equipped with a coil designed for au...
The limited autonomy of flight has long been considered a significant constraint in drone systems. In the context of drone inspections of power lines, this study focuses on a drone equipped with a coil designed for autonomous battery charging. Positioned atop the drone is a charging coil, and on the upper surface, there is an upward-facing camera with a restricted field-of-view (FOV), serving the crucial role of aligning the drone with power lines. This research introduces a wireless power transfer strategy to facilitate the self-charging process of the drone while maintaining its position beneath power lines. To enable the drone's autonomous approach to power lines, a method is devised to guide the drone from below. As the drone approaches the power lines, it monitors the current induced by these lines and gradually adjusts its altitude until a sufficiently high induced current is achieved. The proposed approach ensures safety, as the drone avoids direct contact with high-voltage power lines, setting it apart from other existing methods that require such contact. To the best of our knowledge, this study offers a novel approach to precisely controlling drone movements, and establishing reliable wireless power transfer from power lines. The effectiveness of the developed wireless transfer strategy has been verified through MATLAB simulations.
Graphene PN Junction (GPNJ) logic circuits received significant attention from the researchers thanks to the availability of electrostatically doped graphene PN Junction (GPNJ) device- a promising one for designing lo...
详细信息
Early diagnosis and treatment planning are greatly aided by the early identification of brain tumors. Because of its high resolution, low radiation, and low risk of patient discomfort, magnetic resonance imaging (MRI)...
Early diagnosis and treatment planning are greatly aided by the early identification of brain tumors. Because of its high resolution, low radiation, and low risk of patient discomfort, magnetic resonance imaging (MRI) is frequently used to diagnose brain tumors. Brain tumor identification is only one area where recent developments in convolutional neural networks (CNNs) have shown exceptional effectiveness. This article summarizes recent progress in detecting brain tumors by utilizing MRI images and bespoke CNN layers with transfer learning. The review kicks off with a discussion of the difficulties of detecting brain cancers, such as the tumors' complexity and heterogeneity and the scarcity of available annotated data. The article proceeds to go into the foundations of CNNs and their applicability to MRI image processing. To improve detection accuracy, we incorporate custom CNN layers that are tailored to capture salient tumor-specific information. The concept of transfer learning, in which CNN models trained on large-scale datasets are repurposed for brain tumor detection, is also discussed at length in the review. Using transfer learning, we can take advantage of what we've learned about general image identification to better train models to spot brain tumors. Fine-tuning, feature extraction, and other transfer learning methods are addressed at length. Recent research using custom CNN layers and transfer learning approaches to detect brain cancers in MRI images is thoroughly analyzed in this study. Among the benefits and drawbacks discussed are the methods' adaptability to small datasets, enhanced detection accuracy, and decreased training time. Also, the significance of using metrics for measuring performance and benchmark datasets for comparing methods fairly is discussed. The analysis concludes with suggestions for future study, such as the combination of functional and diffusion tensor imaging with conventional MRI scans to better detect brain tumors. Further
Whenever a read operation is performed in a flash memory storage device, the surrounding cells in the same block are affected and their reliability gradually decreases. This phenomenon is known as read disturb problem...
Whenever a read operation is performed in a flash memory storage device, the surrounding cells in the same block are affected and their reliability gradually decreases. This phenomenon is known as read disturb problem. A remedy to this read disturb is to count the number of read operations for each block and perform read reclaim, the process of moving the existing data to a new block when the count reaches a predefined threshold. However, as the number of blocks in the flash memory device increases, maintaining the per-block read count requires a great amount of memory space in the controller. Therefore, system designers are forced to maintain read counts for a group of blocks (i.e. superblock) which results in the loss of accuracy. This paper proposes novel read disturb management schemes that can reduce the number of read reclaims significantly even if the read count is maintained for each superblock. In the proposed Pointer-based scheme, we keep track of the last block read so as to avoid excessive increase in the read count when the data is read sequentially. We also propose the Bitmap-based scheme that can successfully approximate the actual read count in the presence of random reads, with a negligible space overhead. Our experiments with real-world traces show that the Pointer-based scheme and the Bitmap-based scheme reduce the number of read reclaims by 65.5% and 90.5% on average, respectively, compared to the conventional scheme.
暂无评论