UAV-based object detection has recently attracted a lot of attention due to its diverse applications. Most of the existing convolution neural network based object detection models can perform well in common object detection cases. However, due to the fact that objects in UAV images are spatially distributed in a very dense manner, these methods have limited performance for UAV-based object detection. In this paper, we propose a novel transformer-based object detection model to improve the accuracy of object detection in UAV images. To detect dense objects competently, an advanced foreground enhancement attention Swin Transformer (FEA-Swin) framework is designed by integrating context information into the original backbone of a Swin Transformer. Moreover, to avoid the loss of information of small objects, an improved weighted bidirectional feature pyramid network (BiFPN) is presented by designing the skip connection operation. The proposed method aggregates feature maps from four stages and keeps abundant information of small objects. Specifically, to balance the detection accuracy and efficiency, we introduce an efficient neck of the BiFPN network by removing a redundant network layer. Experimental results on both public datasets and a self-made dataset demonstrate the performance of our method compared to the state-of-the-art methods in terms of detection accuracy.

We analyzed seasonality and interannual variability of tropospheric HCN column amounts in densely populated eastern China for the first time. The results were derived from solar absorption spectra recorded with ground-based high spectral resolution Fourier transform infrared (FTIR) spectrometer at Hefei (117°10′E, 31°54′N) between 2015 and 2018. The tropospheric HCN columns over Hefei, China showed significant seasonal variations with three monthly mean peaks throughout the year. The magnitude of the tropospheric HCN column peak in May > September > December. The tropospheric HCN column reached a maximum of (9.8 ± 0.78) × 1015 molecules/cm2 in May and a minimum of (7.16 ± 0.75) × 1015 molecules/cm2 in November. In most cases, the tropospheric HCN columns at Hefei (32°N) are higher than the FTIR observations at Ny Alesund (79°N), Kiruna (68°N), Bremen (53°N), Jungfraujoch (47°N), Toronto (44°N), Rikubetsu (43°N), Izana (28°N), Mauna Loa (20°N), La Reunion Maido (21°S), Lauder (45°S), and Arrival Heights (78°S) that are affiliated with the Network for Detection of Atmospheric Composition Change (NDACC). Enhancements of the tropospheric HCN columns were observed between September 2015 and July 2016 compared to the counterpart measurements in other years. The magnitude of the enhancement ranges from 5 to 46% with an average of 22%. Enhancement of tropospheric HCN (ΔHCN) is correlated with the coincident enhancement of tropospheric CO (ΔCO), indicating that enhancements of tropospheric CO and HCN were due to the same sources. The GEOS-Chem tagged CO simulation, the global fire maps and the PSCFs (Potential Source Contribution Function) calculated using back trajectories revealed that the seasonal maxima in May is largely due to the influence of biomass burning in South Eastern Asia (SEAS) (41 ± 13.1%), Europe and Boreal Asia (EUBA) (21 ± 9.3%) and Africa (AF) (22 ± 4.7%). The seasonal maxima in September is largely due to the influence of biomass burnings in EUBA (38 ± 11.3%), AF (26 ± 6.7%), SEAS (14 ± 3.3%), and Northern America (NA) (13.8 ± 8.4%). For the seasonal maxima in December, dominant contributions are from AF (36 ± 7.1%), EUBA (21 ± 5.2%), and NA (18.7 ± 5.2%).The tropospheric HCN enhancement between September 2015 and July 2016 at Hefei (32°N) were attributed to an elevated influence of biomass burnings in SEAS, EUBA, and Oceania (OCE) in this period. Particularly, an elevated fire number in OCE in the second half of 2015 dominated the tropospheric HCN enhancement in September – December 2015. An elevated fire number in SEAS in the first half of 2016 dominated the tropospheric HCN enhancement in January – July 2016.