Power line targets (high-voltage lines) are difficult to detect in low-altitude aircraft collision avoidance systems due to their slender physical dimensions and small electromagnetic scattering cross-section, which severely constrains flight safety. Conventional detection methods such as optical radar, millimeter-wave radar, and lidar are limited by factors like visibility, transmission power, and atmospheric conditions, resulting in restricted detection range, low recognition probability, and high false alarm rates. To address these issues, an airborne power line detection system based on a dual-polarization technical scheme is designed. This system utilizes the differential scattering characteristics of power lines under horizontal and vertical polarization in the L-band. It extracts the dual-polarization amplitude and polarization tilt angle from the power line echoes as polarization feature vectors, which are combined with traditional Doppler detection state vectors to form an augmented state vector for target detection. A cascaded classifier, employing support vector machines (SVM) and convolutional neural networks (CNN), is constructed to extract both explicit and implicit features of power line targets. The classifier is trained with 2 000 sets of real detection data to enhance the accurate recognition probability of power line targets in complex ground clutter. Flight tests conducted in field and mountainous environments demonstrate that the proposed power line detection system operates effectively all-weather and all-day, achieving a correct recognition probability of over 92% for power line targets within a range of 1 500 meters, which indicates broad potential for engineering applications