The KIII model is a bionic olfactory model proposed based on the physiological structure of the animal olfactory system and has been applied to pattern recognition problems such as tea classification and EEG recognition. To explore the versatility of the KIII model and improve its performance of the KIII model, this study improved the KIII model by introducing adaptive histogram equalization, Gaussian filtering, and feature fusion methods based on gridded feature extraction and cropping in the preprocessing stage, the effective extraction of image features enhances the model recognition ability; Aiming at the computational cost of solving the dynamic equations of neurons in the KIII model, the fourth-order Runge-Kutta and Euler methods are weighed, and the results show that the Euler method has a relatively low cost in terms of time overhead. Low cost and good recognition results; compared several different measurement algorithms in the decision-making stage, according to the analysis, the Spearman correlation coefficient performed best in this scenario; and was used in the new field of traffic sign recognition, The experimental results show that the improved KIII model has achieved an average recognition effect of 95.18% on the TSRD dataset.