Test, plus the electrical properties of each defect are various to consider the existence of three unique defects inside the same two-dimensional section of the wood. The relative dielectric constants from the 3 defects are 20, 40, and 60, respectively, along with the live wood defect model is set up as shown in Figure 6a, exactly where the Appl. Sci. 2021, 11, x FOR PEER Review 13 of 17 relative dielectric continuous on the defect on the ideal side of your xylem is 20, the relative dielectric continual of your defect above the xylem is 40, plus the relative dielectric continuous with the defect under the xylem is 60. The effect of every single algorithm for defect inversion is shown in dielectric continual of the defect under the xylem is 60. The impact of every single algorithm for Figure 6.defect inversion is shown in Figure six.(a) (b)(c) (d)Figure 6. Heterogeneous multidefect model inversion imaging. (a) Heterogeneous multidefect model with 2cm radius. (b) Figure 6. Heterogeneous multi-defect model inversion imaging. (a) Heterogeneous multi-defect model with two cm radius. CSI inversion benefits. (c) BP neural network inversion results. (d) Modeldriven deep mastering network inversion benefits. (b) CSI inversion results. (c) BP neural network inversion outcomes. (d) Model-driven deep mastering network inversion benefits.As shown in Figure six, for the detection of heterogeneous multidefects inside the As shown in Figure six, for the detection of heterogeneous multi-defects inside the trees, the CSI cannot locate the defect place. The BP neural network better inverts the trees, the CSI can’t locate the defect place. The BP neural network improved inverts the defect size and location, though the boundary involving wood and air inside the result is not defect size and location, even though the boundary in between wood and air within the outcome just isn’t clear clear adequate, plus the IOU values for BP are 0.928 and 0.941, indicating that this algorithm enough, as well as the IOU values for BP are 0.928 and 0.941, indicating that this algorithm is just isn’t correct enough for function extraction from the instruction data. The modeldriven deep studying inversion has significantly less noise, accurately reflecting the defect size and place, and also clearly reflecting the media boundary amongst wood, defect and air, and also the IOU value Alkannin Inhibitor reaches 0.961. As shown in Table 5, under the regular of imply square error, the outcome of the modeldriven depth neural network is drastically greater than that of the BP neuralAppl. Sci. 2021, 11,14 ofnot accurate adequate for feature extraction from the training data. The model-driven deep 1-EBIO In Vivo understanding inversion has less noise, accurately reflecting the defect size and place, and also clearly reflecting the media boundary in between wood, defect and air, and the IOU value reaches 0.961. As shown in Table 5, under the regular of imply square error, the result from the modeldriven depth neural network is significantly superior than that from the BP neural network. The consumption with the two solutions is roughly exactly the same.Table five. Mean square error and typical single detection time for every single algorithm. Contrast Source InversionAppl. Sci. 2021, 11, x FOR PEER Evaluation Mean Square Error Single Detection TimeBP Neural Network 0.2679 0.077 sModel-Driven Deep Learning Networks 0.1345 17 14 of 0.065 sNone None3.6. Algorithm Iterative Stability Analysis three.6. Algorithm Iterative Stability Evaluation BP neural networks plus the model-driven deep lea.