In recent years, digital watermarking has improved the accuracy and resistance of watermarked images against many assaults, such as various noises and random dosage characteristics. Because, based on the most recent assault, all existing watermarking research techniques have an acceptable level of resistance. The deep learning approach is one of the most remarkable methods for guaranteeing maximal resistance in the watermarking system's digital image processing. In the digital watermarking technique, a smaller amount of calculation time with high robustness has recently become a difficult challenge. In this research study, the light weight convolution neural network (LW-CNN) technique is introduced and implemented for the digital watermarking scheme, which has more resilience than any other standard approaches. Because of the LW-CNN framework's feature selection, the calculation time has been reduced. Furthermore, we have demonstrated the robustness of two distinct assaults, collusion and geometric type. This research work has reduced the calculation time and made the system more resistant to current assaults.