Abnormal tumor image identification from brain Magnetic Resonance Images (MRI) is essential for medical diagnostics. In this research, Cascaded Convolutional Neural Network (CCNN) with Haar wavelet features based brain tumor detection technique has been proposed for automatic identification of tumor images. The significant LL sub-band features are first extracted in all image slices. These slices are further processed using CCNN architecture for brain tumor detection. In this architecture, each image slice is convolved with three different 7 x 7, 3 x 3 and 5 x 5 kernels to produce three separate feature maps. These feature maps are cascaded to be processed into the hierarchy of convolutional, pooling and softmax layers to predict whether an image is having a tumor or not. This proposed algorithm is implemented using the BRATS-2018 training dataset. It achieves 96% of Accuracy, 97 % of F1-score, 97 % of Precision, 97 % of Specificity and 96 % of Sensitivity values.