Correction: Kamiya et al. Non-Destructive Estimation of Paper Fiber Using Macro Images: A Comparative Evaluation of Network Architectures and Patch Sizes for Patch-Based Classification. NDT 2024, 2, 487–503

A few unintended typos were introduced after the proofreading stage, and the authors hence wish to make the following corrections to this paper [1]. The authors state that the scientific conclusions are unaffected. This correction was approved by the Academic Editor. The original publication has also been updated.

Error in Table

In Section 4.3 Performance of Text-Containing Images, the values in Table 5 mistakenly showed the same values as Table 3. The corrected

Table 5. Macro image fiber estimation accuracy of text-containing images (%).

Network	500 × 500	750 × 750	1000 × 1000
EfficientNet-B0	60.8	65.1	70.4
DenseNet-201	57.5	65.4	65.4
DarkNet-53	71.6	65.2	61.4
Inception-v3	63.7	68.9	61.2
Xception	65.8	66.3	68.0
Inception-ResNet-v2	67.7	68.6	72.7
NASNet-Large	62.7	64.6	70.0
Average	64.3	66.3	67.0

is as follows:

Text Correction

Due to the correction above, a correction has been made to Section 4. Discussion, Sub-section 4.3. Performance of Text-Containing Images, Paragraph 3:

The results indicated a general decrease in patch classification accuracy for text-containing images in comparison with text-free images. For instance, the accuracy reached 69.1% for text-containing images when Inception-ResNet-v2 was used with 1000 × 1000 pixel patches, which was lower than the accuracy of 82.7% achieved for text-free images. However, the macro image estimation accuracy remained relatively high, with Inception-ResNet-v2 achieving an accuracy of 72.7% for 1000 × 1000 pixel patches; this was comparable to its performance on text-free images. The disparity between patch-level and macro-level performances indicates that the majority voting process used in macro image estimation can partially compensate for the challenges posed by the presence of text in individual patches.