Search Results (8)

Pest infestations remain a critical threat to global agriculture, significantly compromising crop yield and quality. While accurate pest detection forms the foundation of precision pest management, current approaches face two primary challenges: (1) the scarcity of comprehensive multi-scale, multi-category pest datasets and (2) performance limitations in detection models caused by substantial target scale variations and high inter-class morphological similarity. To address these issues, we present three key contributions: First, we introduce Insect25—a novel agricultural pest detection dataset containing 25 distinct pest categories, comprising 18,349 high-resolution images. This dataset specifically addresses scale diversity through multi-resolution acquisition protocols, significantly enriching feature distribution for robust model training. Second, we propose GC-Faster RCNN, an enhanced detection framework integrating a hybrid attention mechanism that synergistically combines channel-wise correlations and spatial dependencies. This dual attention design enables more discriminative feature extraction, which is particularly effective for distinguishing morphologically similar pest species. Third, we implement an optimized training strategy featuring a cosine annealing scheduler with linear warm-up, accelerating model convergence while maintaining training stability. Experiments have shown that compared with the original Faster RCNN model, GC-Faster RCNN has improved the average accuracy mAP0.5 on the Insect25 dataset by 4.5 percentage points, and mAP0.75 by 20.4 percentage points, mAP0.5:0.95 increased by 20.8 percentage points, and the recall rate increased by 16.6 percentage points. In addition, experiments have also shown that the GC-Faster RCNN detection method can reduce interference from multiple scales and high similarity between categories, improving detection performance. Full article

The durability degradation of concrete structures in marine and urban underground environments is largely governed by chloride-induced corrosion. This process becomes significantly more severe under the coupled action of external loading and drying–wetting cycles, which accelerate chloride transport and structural deterioration. However, the existing research often isolates the effects of mechanical loading or environmental exposure, failing to comprehensively capture the synergistic interaction between these factors. This lack of understanding of chloride ingress under simultaneous mechanical and environmental loading limits the development of reliable service life prediction models for concrete structures. In this study, a self-made loading system was employed to simulate this coupled environment, combining external loading with 108 days of drying–wetting cycles. Chloride profiles were obtained to assess the combined effects of stress level, water/binder ratio, and fiber content on chloride penetration in fiber-reinforced cementitious composites (FRCCs). To further extend the analysis, a Crank–Nicolson-based finite difference approach was developed for the numerical assessment of chloride diffusion in concrete structures after repair. This model enables the point-wise treatment of nonlinear chloride concentration profiles and provides space- and time-dependent chloride concentration distributions. The results show that using an FRCC as a repair material significantly enhances the service life of chloride-contaminated concrete structures. The remaining service life of the repaired concrete was extended by 36.82% compared to the unrepaired case, demonstrating the clear practical value of FRCC repairs in aggressive environments. Full article

A well designed transition control schedule can enable the engine to quickly and smoothly transition from one operating state to another, thereby enhancing the maneuverability of the aircraft. Although traditional pointwise optimization methods are fast in solving the transition control schedules, their optimized control schedules suffer from fluctuation problems. While global optimization methods can suppress fluctuation problems, their slow solving speed makes them unsuitable for engineering applications. In this paper, a combined optimization method for the transition control schedules of aero-engines is proposed. This method divided the optimization of the control schedules into two layers. In the outer-layer optimization, the global optimization technique was utilized to suppress the fluctuation of geometrically adjustable parameters. In the inner-layer optimization, the pointwise optimization technique was adopted to quickly obtain the control schedule of fuel flow rate. Moreover, a construction method of non-uniform control points in the global optimization layer was proposed, which significantly reduced the number of control points that needed to be optimized; thus, improving the efficiency of global optimization. The optimization problem of the acceleration and deceleration control schedules of a mixed-flow turbofan engine was used to verify the effectiveness of the combined optimization method. The results show that, compared with the pointwise optimization method, the transition time optimized by the combined optimization method shows no obvious difference. The control schedules optimized by the combined optimization method are not only smooth but can also prevent some components from approaching their working boundaries. Full article

Binary neural networks (BNNs) can substantially accelerate a neural network’s inference time by substituting its costly floating-point arithmetic with bit-wise operations. Nevertheless, state-of-the-art approaches reduce the efficiency of the data flow in the BNN layers by introducing intermediate conversions from 1 to 16/32 bits. We propose a novel training scheme, denoted as BNN-Clip, that can increase the parallelism and data flow of the BNN pipeline; specifically, we introduce a clipping block that reduces the data width from 32 bits to 8. Furthermore, we decrease the internal accumulator size of a binary layer, usually kept using 32 bits to prevent data overflow, with no accuracy loss. Moreover, we propose an optimization of the batch normalization layer that reduces latency and simplifies deployment. Finally, we present an optimized implementation of the binary direct convolution for ARM NEON instruction sets. Our experiments show a consistent inference latency speed-up (up to $\mathbf{1.3}$ and $\mathbf{2.4}\times$ compared to two state-of-the-art BNN frameworks) while reaching an accuracy comparable with state-of-the-art approaches on datasets like CIFAR-10, SVHN, and ImageNet. Full article

The rapid advancement in AI requires efficient accelerators for training on edge devices, which often face challenges related to the high hardware costs of floating-point arithmetic operations. To tackle these problems, efficient floating-point formats inspired by block floating-point (BFP), such as Microsoft Floating Point (MSFP) and FlexBlock (FB), are emerging. However, they have limited dynamic range and precision for the smaller magnitude values within a block due to the shared exponent. This limits the BFP’s ability to train deep neural networks (DNNs) with diverse datasets. This paper introduces the hybrid precision (HPFP) selection algorithms, designed to systematically reduce precision and implement hybrid precision strategies, thereby balancing layer-wise arithmetic operations and data path precision to address the shortcomings of traditional floating-point formats. Reducing the data bit width with HPFP allows more read/write operations from memory per cycle, thereby decreasing off-chip data access and the size of on-chip memories. Unlike traditional reduced precision formats that use BFP for calculating partial sums and accumulating those partial sums in 32-bit Floating Point (FP32), HPFP leads to significant hardware savings by performing all multiply and accumulate operations in reduced floating-point format. For evaluation, two training accelerators for the YOLOv2-Tiny model were developed, employing distinct mixed precision strategies, and their performance was benchmarked against an accelerator utilizing a conventional brain floating point of 16 bits (Bfloat16). The HPFP selection, employing 10 bits for the data path of all layers and for the arithmetic of layers requiring low precision, along with 12 bits for layers requiring higher precision, results in a 49.4% reduction in energy consumption and a 37.5% decrease in memory access. This is achieved with only a marginal mean Average Precision (mAP) degradation of 0.8% when compared to an accelerator based on Bfloat16. This comparison demonstrates that the proposed accelerator based on HPFP can be an efficient approach to designing compact and low-power accelerators without sacrificing accuracy. Full article

Wind power utilization is attracting worldwide attention in the renewable energy field, and as wind power develops from land to sea, the size of the blades is becoming incredibly larger. The fatigue test, especially the biaxial synchronous fatigue test for the blades, is becoming an indispensable step to ensure the blade’s quality before mass production, which means the biaxial independent test presently used may have difficulty reproducing the real damage for large-sized blades that oscillate simultaneously in flap-wise and edgewise directions in service conditions. The main point of the fatigue test is to carry out accelerated and reinforced oscillations on blades in the experimental plan. The target moments of critical blade sections are reached or not during the test are treated as one significant evaluation criterion. For independent tests, it is not hard to realize moment matching using additional masses fixed on certain critical blade sections, which may be not easy to put into effect for biaxial synchronous tests, since the mechanical properties and target moments in the flap-wise and edgewise directions are widely varied. To realize the mechanical decoupling for loading force or additional mass inertia force in two directions is becoming one of the key issues for blade biaxial synchronous fatigue testing. For this problem, the present paper proposed one mechanical decoupling design concept after a related literature review. After that, the blade moment design and target matching approach are also proposed, using the Transfer Matrix Method (TMM) for moment quick calculation and Particle Swarm Optimization (PSO) for case optimization. Full article

GPS-based, satellite-to-satellite tracking observations have been extensively used to elaborate the long-scale features of the Earth’s gravity field from dedicated satellite gravity missions. We proposed compiling a satellite gravity field model from Gravity Field and Steady-State Ocean Circulation Explorer (GOCE) satellite accelerations directly estimated from the onboard GPS data using the point-wise acceleration approach, known as the carrier phase differentiation method. First, we composed the phase accelerations from the onboard carrier phase observations based on the sixth-order seven-point differentiator, which can eliminate the carrier phase ambiguity for Low Earth Orbiter (LEO). Next, the three-dimensional (3D) accelerations of the GOCE satellite were estimated from the derived phase accelerations as well as GPS satellite ephemeris and precise clock products. Finally, a global gravity field model up to the degree and order (d/o) 130 was compiled from the 71 days and nearly 2.5 years of 3D satellite accelerations. We also recovered three gravity field models up to d/o 130 from the accelerations derived by differentiating the kinematic orbits of European Space Agency (ESA), Graz, and School of Geodesy and Geomatics (SGG), which was the orbit differentiation method. We analyzed the accuracies of the derived accelerations and the recovered gravity field models based on the carrier phase differentiation method and orbit differentiation method in time, frequency, and spatial domain. The results showed that the carrier phase derived acceleration observations had better accuracy than those derived from kinematic orbits. The accuracy of the recovered gravity field model based on the carrier phase differentiation method using 2.5 years observations was higher than that of the orbit differentiation solutions for degrees greater than 70, and worse than Graz-orbit solution for degrees less than 70. The cumulative geoid height errors of carrier phase, ESA-orbit, and Graz-orbit solutions up to degree and order 130 were 17.70cm, 21.43 cm, and 22.11 cm, respectively. Full article

To accelerate photoacoustic data acquisition, in [R. Nuster, G. Zangerl, M. Haltmeier, G. Paltauf (2010). Full field detection in photoacoustic tomography. Optics express, 18(6), 6288–6299] a novel measurement and reconstruction approach has been proposed, where the measured data consist of projections of the full 3D acoustic pressure distribution at a certain time instant T. Existing reconstruction algorithms for this kind of setup assume a constant speed of sound. This assumption is not always met in practice and thus can lead to erroneous reconstructions. In this paper, we present a two-step reconstruction method for full field detection photoacoustic tomography that takes variable speed of sound into account. In the first step, by applying the inverse Radon transform, the pressure distribution at the measurement time is reconstructed point-wise from the projection data. In the second step, a final time wave inversion problem is solved where the initial pressure distribution is recovered from the known pressure distribution at time T. We derive an iterative solution approach for the final time wave inversion problem and compute the required adjoint operator. Moreover, as the main result of this paper, we derive its uniqueness and stability. Our numerical results demonstrate that the proposed reconstruction scheme is fast and stable, and that ignoring sound speed variations significantly degrades the reconstruction. Full article