Airway management is one of the most important priorities when dealing with patients with severe injuries, but knowledge of the important anatomy and physiology is needed for providers to perform a successful surgery. This paper provides a solution for the precise cricothyroid membrane detection problem for real-time surgical airway management applications. With a commercial compact and portable cricothyrotomy kit, the proposed method will enable providers with general knowledge to perform successful first-aid airway management. In this paper, we propose a Hybrid Neural Network (HNNet), consisting of two parallel computing ensembles. The first ensemble takes as an input a low-resolution global image and outputs the Region-of-Interest (ROI) from the predefined grids. The high-resolution image is then cropped according to the ROI, and fed into the second ensemble to achieve precise keypoint detection. Global features and their spatial information from the first ensemble are also fed into the second ensemble to improve the precision. A dataset that consists of over 16,000 images from 13 subjects is built, and the location of the cricothyroid membrane in each image is precisely labeled by medical experts. The training results are presented to show both the efficiency and improved performance of our proposed method compared to existing ones.