The exchange of text messages and media files in modern digital communication demands an end-to-end encryption (E2EE) scheme that simultaneously guarantees confidentiality, integrity, authenticity, forward secrecy, and resilience to nonce mismanagement. The widely deployed AES-GCM cipher is known to collapse catastrophically under nonce reuse, enabling plaintext recovery via ciphertext XOR and universal authentication tag forgery, while prior studies have not integrated AES-GCM-SIV, ECDH Curve25519, and HKDF-SHA256 into a unified scheme covering both text and media with comprehensive quantitative measurement. This study aims to design, implement, and evaluate an E2EE scheme based on an ECDH Curve25519, HKDF-SHA256, and AES-256-GCM-SIV pipeline across security and performance dimensions. A quantitative comparative experimental approach was employed using a Rust-based simulation program with 30 iterations per algorithm–data combination across twelve test categories (text of 100–2,000 bytes; image, audio, and video files up to 33.69 MB), accompanied by five security scenarios under the Dolev–Yao threat model (eavesdropping, tampering, replay attack, nonce reuse, and Man-in-the-Middle) and an ablation study of each cryptographic component. The results demonstrate that the proposed scheme satisfies every claimed security property within the defined threat model and tested scenarios, whereas the AES-GCM baseline fails the nonce-reuse scenario. On small data, AES-GCM-SIV incurs a 33.7%–51.6% encryption-time overhead; yet on media files it outperforms the baseline with 11.3%–16.9% higher throughput (peaking at 14.8% on large video), while memory consumption and size overhead remain identical. The ablation study confirms that each component delivers an irreplaceable security contribution. This research contributes empirical evidence that the combination of ECDH Curve25519, HKDF-SHA256, and AES-256-GCM-SIV is a viable cryptographic pipeline for E2EE simulation schemes resilient to implementation errors.
Accurate storage performance characterization in virtualized environments requires benchmarking tools with minimal runtime overhead. Conventional asynchronous I/O introduces scheduling overhead and latency variability, degrading measurement stability. This study designs and evaluates Calyx, a deterministic disk benchmarking tool built on Linux io_uring that integrates submission queue polling, fixed resource registration, and an allocation-free I/O path. Evaluation on HyperV virtualized NVMe compared Calyx against fio. Under sequential reads on ext4, Calyx achieves 4,576.92 MB/s throughput with 0.18% CV and 3,979.54 μs P99 latency, versus fio's 4,560.49 MB/s, 0.73% CV, and 5,210.11 μs P99 latency. For sequential writes on XFS, Calyx maintains 3,320.09 MB/s (2.25% CV, 9,529.98 μs P99) while fio drops to 1,388.92 MB/s (49.57% CV, 39,059.46 μs P99), which is plausibly attributed to SSD cache exhaustion. CPU utilization for Calyx scales flatly from 49.5% (QD=8) to 53.1% (QD=64), whereas fio scales linearly from 35.2% to 59.8%. Calyx's deterministic pipeline reduces measurement variability without sacrificing throughput, confirming that registered-resource execution improves benchmarking reliability in virtualized systems.
Bean leaf disease classification remains challenging due to limited annotated datasets and high visual variability in lesion size, color intensity, illumination, leaf orientation, and background conditions. Although convolutional neural networks and transfer learning have shown promising results, many existing bean leaf disease classification methods still rely on single-stage fine-tuning and end-to-end softmax classification, which may reduce generalization on small agricultural datasets. Furthermore, the combined use of staged transfer learning, MixUp regularization, multi-level deep feature extraction, and classical machine learning classifiers remains insufficiently explored. This gap highlights the need for a more robust and validation-controlled framework to improve classification performance under limited-data conditions. Therefore, this study proposes a hybrid classification framework that integrates two-phase VGG19 transfer learning, validation-selected MixUp regularization, multi-level deep feature extraction, and classical machine learning classifiers. In the first phase, the pretrained VGG19 backbone is frozen to train a tasks-pecific classification head, while in the second phase, selected deeper layers are fine-tuned to adapt high-level representations to bean leaf disease patterns. Deep features extracted from multiple representation levels are classified using SVM, KNN, and LightGBM under a strict validation-based model selection protocol. Experiments on the public IBean dataset show that the proposed hybrid approach outperforms the end-to-end CNN baseline. The best configuration, combining two-phase transfer learning, MixUp regularization, and LightGBM, achieves 98.44% test accuracy on the IBean dataset. Additional validation on an independent Bean Leaf Disease Dataset achieves 98.33% test accuracy. These findings indicate that the proposed framework is effective for improving bean leaf disease classification performance.
Multi-object tracking (MOT) in crowded scenes is difficult because objects frequently occlude each other and share similar appearances, leading to identity confusion. End to-end trackers based on the From Detection to Association (FDTA) architecture reduce inter-object embedding errors but remain susceptible to ID Switches (IDSW) when ID confidence falls briefly during crowd traversal. This paper proposes a training-free density-aware association module that operates on top of a frozen FDTA model at inference time. Identity recovery activates only when both local density and frame-level detection count exceed fixed thresholds, preventing false merges in sparse regions. The method also includes a geometry-based offline tracklet linker that reconnects fragmented trajectories using bounding box constraints alone, with no learned components. On the DanceTrack validation set, the approach reduces IDSW by 13% and raises Higher Order Tracking Accuracy (HOTA) from 64.36 to 65.39, Association Accuracy (AssA) to 57.15, and ID F1 Score (IDF1) to 71.47. On the official test server, the method scores 72.01 HOTA and 78.21 IDF1
Marine Heatwaves (MHW) are extreme sea surface temperature events that significantly affect marine ecosystems, fisheries, and coral reef environments, highlighting the need for accurate forecasting systems in vulnerable regions such as the Lesser Sunda Islands. This study aims to develop a hybrid deep learning-based MHW prediction system by integrating U-Net and ConvLSTM models over the regions of Bali, West Nusa Tenggara, and East Nusa Tenggara. The datasets used consist of NOAA OISST V2.1 daily sea surface temperature data for the period 1985–2024, along with Niño 3.4 and Dipole Mode Index (DMI) atmospheric indices. The U-Net model was applied to predict Sea Surface Temperature Anomaly (SSTA) intensity, while ConvLSTM was used to estimate the probability of MHW occurrence. The results demonstrate that the best forecasting accuracies of the U-Net intensity model for 1-, 3-, 5-, and 7-day lead times reached 0.8896, 0.8597, 0.8462, and 0.8674, respectively, with optimal thresholds of 0.92 °C, 0.56 °C, 0.51 °C, and 0.31 °C. The ConvLSTM probability model produced maximum FAR values of 0.9101, 0.8962, 0.8941, and 0.8854 with optimal probability thresholds of 0.25, 0.1, 0.1, and 0.1. RMSE evaluation increased gradually from 0.27 °C on day-1 to 0.94 °C on day-7 forecasts. Overall, the hybrid deep learning framework demonstrated robust and stable performance in representing both the intensity and probability of MHW events up to a seven-day forecasting horizon.
