Abstract: ‍ ‍With the wide application of drones in various fields， the threats they bring to countries and regions are increasing daily， and effective warnings and countermeasures are imminent. With advancements in drone technology， the difficulty of drone warning and countermeasures is also increasing. Compared to traditional radar， ubiquitous radar uses a wide transmission and narrow reception beam design， utilizing long-time accumulation techniques to obtain sufficient processing gain， thus detecting weak signals of “low， slow， and small” targets in complex clutter backgrounds. Based on high detection sensitivity， ubiquitous radar provides accurate target motion characteristics and high-resolution target Doppler characteristics. By utilizing these features， it is possible to detect and classify drone targets and then use other types of countermeasures to neutralize them effectively. Due to the similarity in the motion trajectory and maneuverability of birds and drones， effectively realizing the classification and identification of drones and birds is a typical problem faced by ubiquitous radar. Based on the ubiquitous radar trajectory Doppler data and real-time classification scenarios， this study proposes an optimized classification workflow based on sequential feature extraction， which includes sliding window feature extraction and real-time classification based on the sliding window. When applied to a system， this process can output the track point class labels in real-time as the target trajectory extends. The core of the real-time classification process is the sliding window-based feature extraction method， which reduces the data dimensionality of feature extraction and increases the probability of similarity between the two types of targets within the sequential window. This requires the features extracted from low-dimensional data to reflect the characteristics of the targets well. Based on the real-time classification problem， this study designs six features， including the correlation coefficient of adjacent window velocities， to describe the degree of velocity change， velocity change stability， and trajectory change stability within the sequential window. The feature classification significance analysis shows that apart from the velocity standard deviation， the distribution of other features extracted based on the sliding window has low overlap. Based on the above methods， this study uses field-measured data from ubiquitous radar for experiments to select the optimal classifier and optimal sequential window size. Random sampling experiments are then used to simulate the real-time operation of the radar system. The overall classification accuracy reaches 92%， and under the optimal window size， the classification accuracy of each track of the two types of targets changes stably. Finally， the classification results are displayed on Amap. The display results show that the drone trajectories and motion changes used in this study are diverse， and the dataset is good in terms of diversity. Experimental verification demonstrates that the proposed features are reasonable and effective and that the real-time classification process is feasible.