This paper proposes a novel pollination process-based, bio-inspired, multi-population global optimization algorithm. The plants optimize their resources spent on floral display, fragrance, pollen, and nectar production to attract pollinators such as bees, insects, and birds etc. Corresponding to a given resource cost, plants reduce their resource expenditure (cost) if the reproduction success is good. However, if the reproduction success is poor, plants increase their cost on floral display, fragrance, and superior nectar contents such that the number of pollinators and revisitation by pollinators increases to improve reproduction success. The proposed pollination-based optimization (PBO) algorithm was evaluated on the 80 test functions of CEC 2021, and the performance was compared with 8 recent algorithms. The algorithm performed exceptionally well, leading in 41 of the 80 functions of the test bench. The paper further, demonstrates the application of the proposed algorithm to evolve an optimized CNN architecture for the rice disease detection from the rice leaf dataset. The rice leaf dataset has 5932 infected images indicating various diseases. The PBO-based approach with 99.37% accuracy outperformed KNN, SVM, Decision Tree, Random Forest, GA-CNN, and BBBC-CNN based algorithms.