We developed a convolutional neural network-based high-sensitivity gas optical sensor (CNN-HGS) using photonic crystals for detecting different gases (N₂, He, and CO₂). Square-shaped dielectric rods with a cubic lattice index were used for the maintenance of the exact distance and area between the rods. The flower pollination optimization algorithm was used for the improvement of the hyperparameters of the CNN. The proposed CNN-HGS consists of three resonance nanocavities: one nanocavity with a refractive index of 2.6 located at the center and the other two subnanocavities with a refractive index of 2.1 placed in the path of the input and output waveguides, respectively. The experiment for the proposed CNN-HGS was conducted with three certified sample gas cylinders and a precise mass flow controller. A clean air generator was linked to the same pipeline for the dilution of the flow of hydrocarbon. During the experiment, a typical gas admixture with concentrations of 60 to 900 ppm was used. Training and testing were conducted using a dataset with 80% and 20% of the total, respectively. The result shows that the proposed optimized CNN delivers a 7.5% improvement in the training accuracy. The CNN-HGS was tested and compared with three other optimization techniques, and the result shows the superiority of the proposed method.