Computer Vision (CV) is constantly inundated with massive volumes of data. One of the most challenging types of data for an Artificial Intelligence (AI) system is imagery data. Convolutional neural networks (CNNs) are utilized to cope with Big Data of such type, but progress is gradual. The 3 Parent Genetic Algorithm (3PGA), an evolutionary computation method, is employed to evolve a default CNN in this study. 3PGA is an extension of GA which has been developed further for better optimization. We observed from the literature that 3PGA is giving excellent results on standard benchmark functions as compared to other recent soft-computing-based approaches. The accuracy of the evolved CNN increased from 53% to 75%, resulting in a net improvement of more than 40%. Furthermore, it was noted that the hyperparametric combinations or features of a CNN, which are very distinct from those commonly utilized, appear to perform better. A geographical landmarks dataset from Google was used for testing purposes. Landmark recognition is one of the most time-consuming jobs for an AI system, and the optimization of a network on a landmarks dataset shows that evolutionary computation can be substantially used in the future for the evolution of Artificial Neural Networks (ANNs).