*3.4. Fusion Model Analysis*

Figure 14 shows the Gibbs sampling dynamics of the health parameters under the condition of tomato leaves infected with leaf mildew. Figure 14a represents the frequency of tomato leaves infected with leaf mildew, while Figure 14b,c each represent a health parameter map of a hyperspectral THz characteristic band.

**Figure 14.** Gibbs sampling diagram. (**a**) represents the frequency of tomato leaves infected with leaf mildew, while (**b**,**c**) each represent a health parameter map of a hyperspectral THz characteristic band.

In Figure 15, a probability density diagram was used to characterize the leaf health parameters of tomato leaf mold. Type I information fusion refers to THz spectral absorbance feature band fusion, while type II information fusion refers to THz spectral power spectrum feature band fusion, and type III information fusion refers to hyperspectral feature band fusion. These three types of information are fused to re-evaluate the health parameter indicators and calculate the recognition rate. After fusing the three types of prior information, it can be seen from the figure that the estimation results were significantly improved after fusing type I information. The posterior distribution of tomato pests and diseased leaves illustrates this point more clearly. The health parameters of tomato leaf mildew posterior samples were also all distributed around 1.75, indicating that the modified Bayesian network model is effective in identifying tomato leaf mildew samples. After the fusion of the prior information, the variables and the actual values increased in agreement, and the final obtained health parameters and posterior distribution of tomato leaves in the state of infection with pests and disease were very close to the actual values.

As shown in Table 4, the overall recognition rate of the improved Bayesian inference for tomato leaf mildew was finally obtained as 97.12%. Therefore, the hyperspectral fusion THz-based technique is feasible for application in tomato leaf mildew recognition.

**Table 4.** Prediction accuracy of each model.


(**a**)

(**b**) **Figure 15.** Schematic diagram of the probability density of health parameters. (**a**) Posterior distribution of changes in health parameters, (**b**) posterior distribution of changes in health parameters after
