Insulation has been widely used in many industrial applications such as Thermal Barrier Coatings and Glass Fibre Reinforced Polymer. The performance of the insulation coatings materials is affected by various defects due to aging and cyclic process. If not immediately detected and repaired, such flaws could lead to catastrophic failure. Therefore, defect detection with microwave NDT is needed in the industry to obtain the result of reflected signals from microwave and determine the shape of defects of the products. This research aims to enhance the deep learning algorithm for defect detection using hyperparameters tuning approaches for the microwave technique. The hyperparameters include learning rate, dropout rate, hidden layers and number of filters. The enhanced deep learning algorithm is compared against the employed algorithm on defect detection in term of performance. The data is preprocessed to get four kinds of data: labelled set for training, input set for training, labelled set for testing and input set for testing. The training and testing process is done using Convolutional Neural Network (CNN). The accuracy of the defect detection by CNN is 91.8% which the defect is not well predicted. After hyperparameter tuning, the accuracy is increased to 94.67%. The predicted defect size is close to the actual defect size.