Cloud plays a crucial role in the Earth's radiation balance and the water cycle, and its formation and evolution are very closely related to weather change. The results of continuous observation and analysis of cloud parameter data can be used in the fields of solar energy production forecasting and meteorological research for aviation and shipping. Cloud cover is one of the main macroscopic parameters of cloud and one of the important elements of ground-based meteorological observations. Thus, the study of its observation methods and approaches has important application value and wide application prospects.

Cloud observation can be divided into air-based observation, space-based observation, and ground-based observation according to the different observation platforms. Among them, ground-based cloud observation has a wide range of fields, low observation cost, and other characteristics, which attracts extensive attention. In the early days, ground-based observation mainly relies on manual observation, which is subjective and has poor continuity. With the continuous development of hardware sensor technology, digital image processing technology, computer technology, and other technologies as well as the increase in the business needs for ground-based cloud observation automation, a series of ground-based cloud observation equipment has been developed.

According to the imaging band of the observation equipment, ground-based cloud observation equipment is mainly divided into two categories for visible band (450-650 nm) and infrared band (8-14 μm) observation. The observation equipment for the visible band can be used only in the daytime and is equipped with a solar baffle or a solar tracking device in most cases. As a result, the acquired sky image is partially blocked, which affects the inversion accuracy of cloud parameters. Regarding the observation equipment for the infrared band, the field of view of the sky observed by one imaging is usually small, and acquiring all-sky images needs scanning and stitching. However, scanning takes a long time, and the movement of clouds causes stitching errors. In addition, scanning and stitching increase system complexity. Therefore, it is important to solve the current problems of ground-based cloud observation equipment so that it can meet the needs for operational ground-based cloud observation.

To realize all-sky cloud cover observation, this paper proposes a dual-band all-sky cloud cover observation system. The visible light imaging unit is based on a fisheye lens and a high-resolution industrial camera and adopts the built-in chopper plate design idea to achieve unobstructed all-sky cloud map acquisition. In addition, through the acquisition of multiple sky images with different exposure values, combined with high dynamic range (HDR) image synthesis technology to synthesize an all-sky image, the impact of overexposure of pixels caused by the sun on imaging can be reduced to a minimum. The infrared imaging unit uses a large array of infrared detectors with a customized infrared wide-angle lens, and a field of view of more than 160°can be observed in a single time. It is faster and easier than the traditional scanning modes which splice images with very limited fields of view and achieves the largest field of view in a single observation among similar equipment. The dual-band all-sky cloud cover observation system works in the way of dual-band observation in the daytime and infrared observation in the nighttime, which thereby realizes all-day cloud observation. To obtain all-sky cloud cover information, the equipment has different built-in cloud cover segmentation algorithms based on the all-sky imaging principle and imaging characteristics of different wavelength bands. For the visible band, a visible cloud image segmentation network, SegCloud, is proposed in light of deep learning technology, while for the infrared band, an infrared cloud segmentation algorithm based on numerical simulation of the infrared raw grayscale image is proposed. To quantitatively analyze the effectiveness of the algorithm, this paper analyzes the consistency and accuracy of the dual-band cloud observation data and verifies the high accuracy of the system in cloud cover observation.

The dual-band all-sky cloud cover observation system proposed in this paper effectively solves the problems of the blocking of traditional visible all-sky imaging equipment and the small observation field of view of the infrared equipment, laying the foundation for the accuracy improvement of ground-based cloud observation. In addition, the dual-band all-sky cloud cover observation design not only enables all-day ground-based cloud observation but also provides multi-channel raw data for the inversion of parameters such as cloud optical thickness, cloud base height, and precipitable water.