<p>The fata morgana algorithm (FATA) is a novel swarm intelligence optimization algorithm, and its design inspiration stems from the unique natural phenomenon of mirages. To overcome the premature convergence of the FATA, which leads to its entrapment in local optimal solutions, an improved FATA (IMFATA) is proposed in this paper based on a circle chaotic map and adaptivee t-Distribution perturbation mutation. The IMFATA is validated against the FATA on 23 benchmark functions and CEC2021 test functions and compared with the honey badger algorithm (HBA), beluga whale optimization (BWO), the whale optimization algorithm (WOA),the goose algorithm(GOOSE), the dung beetle optimizer (DBO), and the aquila optimizer (AO). The experimental results show that the IMFATA can effectively improve its computational accuracy and convergence speed, and its global optimization ability is superior to that of the other algorithms. Finally, the IMFATA is applied to optimize the parameters of a support vector machine (SVM) for Dendrobium grade classification. The experimental results show that the classification accuracy (F1 value) achieved for the Dendrobium grades optimized by the IMFATA is relatively high, fully demonstrating the superiority of the IMFATA in practical engineering applications.</p>