In order to solve the problem of algorithm robustness caused by scale change and imbalanced distribution of classes in the scene text detection task, we propose a new two-stream fusion framework TSFnet. It is constructed by the Detection Stream, the Judge Stream and the Merge output algorithm. In the Detection Stream, we propose a loss balance factor (LBF), which is used to optimize region proposal network. Then, the Regression-net and the Segmentation-net are used to predict text global segmentation map and its corresponding coordinates probability score. In Judge Stream, we use the feature pyramid network to extract the Judge map. In the process, the LBF is calculated to support the Detection Stream. Finally, we design a novel algorithm to fuse the outputs of the two-stream, and the precise position of the text is localized. Extensive experiments are conducted on the ICDAR 2015 and the ICDAR2017-MLT standard datasets. The results demonstrate that the framework performance is comparable with the sate-of-the-art approaches.
16 October 2025 in Journal of King Saud University Computer and Information Sciences
With the growing emphasis on intelligent and sustainable manufacturing, the flexible job shop scheduling problem (FJSP) has emerged as a critical optimization challenge due to its capability to model complex and dynamic production environments. As an NP-hard problem, FJSP poses significantly greater complexity than the classical job shop scheduling problem, particularly in high-precision domains such as aerospace manufacturing, where traditional optimization methods often fall short. To overcome the limitations of the standard Whale Optimization Algorithm (WOA)–namely its continuous search paradigm and inadequate local search mechanisms–this paper proposes a genetic-operator-integrated whale optimization algorithm (GOI-WOA) tailored for discrete scheduling tasks. The GOI-WOA employs a two-layer discrete encoding strategy to represent operations and machine assignments, and introduces genetic operators, a critical-path-based neighborhood search, and a temporal-entropy-based local search to enhance solution diversity and local exploitation. Furthermore, an elite information-sharing mechanism is incorporated to facilitate the diffusion of high-quality solutions within the population. Comprehensive experiments conducted on standard benchmark instances demonstrate that GOI-WOA consistently outperforms baseline WOA and other state-of-the-art methods in terms of makespan minimization, convergence stability, and computational efficiency. These results validate the effectiveness and robustness of GOI-WOA as a competitive solution framework for complex manufacturing scheduling problems.
Aviation security inspection needs lots of time and human labour. In this paper, we establish a new network for detecting prohibited objects in aeronautic security inspection X-ray images. Objects in the X-ray image often present rotated shapes and overlap heavily with each other. In order to solve the rotation and occlusion in X-ray image detection, we construct the de-rotation-and-occlusion module (DROM), an efficient module that can be embedded into most deep learning detectors. Our DROM leverages the edge, colour and oriented fast and rotated brief (ORB) features to generate an integrated feature map, while the ORB features could be extracted quickly and diminish the deviation produced by rotation effectively. Finally, we evaluate DROM on the OPIXRay dataset; compared with several latest approaches, the experimental results certify that our module promotes the performance of single shot multibox detector (SSD) and obtains higher accuracy, which proves the module's application value in practical security inspection.
15 December 2023 in International Journal of Pattern Recognition and Artificial Intelligence