Utilizing machine learning for classifying radar antenna radiation patterns is a novel and significant area in radar signal processing and machine learning. Today, various radar systems for specific applications emit and rotate electromagnetic waves according to their radiation patterns to cover 360 degrees in different angles of the surrounding space. The transmitted waves in the operational environment are received by radar receivers, which are horizontally aligned with the radar. By collecting these waves, the radiation pattern of the transmitting radar antenna can be identified using machine learning algorithms. For this purpose, PDW data related to the radar is used. In this paper, images are created from the PA behavior of the transmitted radar signal in a two-dimensional plot of pulse DOA and PA. Consequently, feature extraction is performed from the radar signal behavior in the image, up to the last layer. Then, the classification of radars based on their application type is done using machine learning. The purpose of this paper is to classify the radiation pattern of the radar antenna with acceptable accuracy by using a suitable convolutional neural network algorithm and generating appropriate data. Generally, the goal is to prove that receiving PDW data and converting it into color images provides us with new input data that can be used as input to the network, and new features can be obtained. New features are defined and added in the radar radiation pattern in terms of the presence of white Gaussian noise, multipath, missing data, and the superposition of two radiation patterns, in order to recognize the degraded radiation pattern appropriate to the operational environment. This paper demonstrates that it can classify the radar antenna radiation pattern with an accuracy of 96.6% in the presence of white Gaussian noise, multipath, missing data, and the superposition of two radiation patterns.