Learning super-resolution and pyramidal convolution residual network for vehicle re-identification

Scientific Reports, Год журнала: 2024, Номер 14(1)

Опубликована: Ноя. 3, 2024

Vehicle re-identification (Vehicle Re-ID) aims at retrieving and tracking the specified target vehicle with multiple other cameras, which can provide help in checking violations catching fugitives, but there are still following problems that need to be solved urgently. First, existing collected Re-ID data often have low resolution blur local regions, so algorithm cannot accurately extract subtle feature representations. In addition, small features easy cause disappearance of under operation a large convolution kernel, makes model unable capture learn features, resulting inaccurate judgment vehicles. this study, we propose method based on super pyramidal residual network. Firstly, super-resolution image generation network leveraging generative adversarial networks (GANs) is proposed. This employs both content loss as optimization criteria, ensuring an efficient transformation from low-resolution into counterpart, while meticulously preserving intricate high-frequency details. Then, multi levels operations designed generate multi-scale information different scales. Moreover, concept learning applied between expedite enhance recognition capabilities. Ultimately, double convolutions employed original image, yielding low-noise representations semantic respectively. By seamlessly fusing these two diverse sources information, resultant combined exhibit heightened discrimination capabilities significantly bolster robustness features. order verify effectiveness proposed method, extensive experiments carried out VeRi-776 VehicleID datasets. The experimental results show paper effectively captures detail images, distinguishes differences vehicles same type, superior state-of-the-art methods.

Язык: Английский

Prominent involvement of acetylcholine in shaping stable olfactory representation across the Drosophila brain

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Апрель 5, 2024

Despite the vital role of neuromodulation in neural system, specific spatiotemporal dynamics neuromodulators and their interactions with neuronal activities vivo are still unclear, hampering our understanding information representation functional contributions systemically. To address this problem, we employed two-photon synthetic aperture microscopy (2pSAM) to record long-term neuromodulatory olfactory responses across Drosophila brain at high speed. Our results revealed distinct response properties, global propagation, connectivity, odor identity among neuronal, cholinergic, serotoninergic multiple regions. We discovered compensation between activity cholinergic dynamics, both connectivity network structures Moreover, employing low-dimensional manifold analyses, characterized stable over a long term. Collectively, unbiased comprehensive investigation unveiled prominent involvement acetylcholine (ACh) shaping brain, underscoring inadequacy solely considering when examining brain.

Язык: Английский