Classification of the radar targets using efficient convolution neural networks is proposed. It uses the time-frequency image features of the target for classification. Machine learning is an extremely time-consuming process as the feature extraction of the image needs strong knowledge of the subject and domain. To overcome this, deep learning is used to automate the process of feature extraction. The simple and complex targets are modelled using a dedicated machine learning algorithm, and their respective spectrogram images are created. These are partitioned into training and testing data sets for training the CNN (alexnet) and hybrid models of CNN where the feature extraction is done using CNN and classifying the target is done using machine learning algorithms like SVM, KNN. A frequency domain filter using CNN is used to pre-process the time-frequency image features for frequency smoothing and dimension reduction. The hybrid model of CNN with an SVM classifier resulted in high accuracy with good performance.