Wood board edge sealing detection experiment method based on improved YOLOv8

Abstract

The invention discloses an improved YOLOv8-based board edge sealing detection experiment method. The method comprises the following steps: S10, preparing an experiment environment; s20, evaluation index selection; s30, performing an ablation experiment; s40, performing a model contrast experiment; and S50, carrying out an attention mechanism contrast experiment. According to the invention, the average detection precision mAP50 and mAP50: 95 are respectively improved by 2.4% and 1.3%, the FPS is also small in difference, and the requirement of real-time detection of the edge sealing defect of the wood board can be met. Compared with other mainstream detection algorithms at present, the method has the advantage in average detection precision, and shows the effectiveness of the improved model.

Description

Technical Field

[0001] The invention belongs to the field of image feature fusion, and in particular relates to a wood board edge banding detection experimental method based on improved YOLOv8. Background Art

[0002] IntroductionWith the economic development and social progress of the times, residents have more and more demands for furniture, and the demand for wooden furniture is growing steadily every year. Wooden furniture is popular among people because of its advantages of environmental protection, health, beauty, durability, and high cost performance. The large consumption of wooden furniture brings profits to furniture manufacturers, but also brings higher requirements for the product quality of furniture. Wood board is the basis for making furniture and is also a crucial link affecting the quality of wooden furniture. Wood board defect detection includes multiple steps such as defect identification, defect classification, and defect location. Among them, edge banding is an important part of wood board. Good quality edge banding can make the board more beautiful, reduce the erosion of air impurities on the inside of the board and the release of formaldehyde, which is beneficial to environmental protection. In the production process, various factors such as manufacturing process and board material can cause edge banding defects. The increasing quality requirements of board put forward strict requirements for board manufacturers in the production and quality inspection of board. However, at present, many wood board production lines still use manual means to detect board edge banding defects and their quality. Because manual detection methods have the disadvantages of low detection efficiency, high false detection and missed detection rates, furniture manufacturers need better means to improve the ability of board edge banding detection. Therefore, it is of great significance to study an automated detection method for board edge banding defects.

[0003] The existing technology for wood board edge banding detection has the problems of small data sets and insufficient defect types. The current deep learning has low detection rate and missed detection in wood board edge banding defect detection, especially when there are a large number of targets. Summary of the invention

[0004] In view of this, the present invention proposes a wood board edge detection experimental method based on improved YOLOv8, comprising the following steps:

[0005] S10, experimental environment preparation;

[0006] S20, selection of evaluation indicators;

[0007] S30, ablation experiment;

[0008] S40, model comparison experiment;

[0009] S50, attention mechanism comparison experiment.

[0010] Preferably, the experimental environment preparation in S10 includes establishing a data set and setting training parameters.

[0011] Preferably, the established data set includes at least 6,000 images of wood board edge banding defects, and the defect types include: debonding, short tape, residual glue, chipping, glue seam, excessive length, residual tape and dirt. The data set is divided into a training set, a validation set and a test set in a ratio of 8:1:1.

[0012] Preferably, the training parameters include: initial learning rate 0.01, final learning rate 0.01, number of threads 8, weight decay coefficient 0.0005, momentum 0.937, batch size 16 and number of iterations 250.

[0013] Preferably, the evaluation indicators in S20 include: recall rate R, mean average precision mAP50 and mAP50:95.

[0014] Preferably, the ablation experiment in S30 includes 5 groups, namely, a YOLOv8 structure, a BiFPN feature pyramid network, a small target layer, a BiFPN feature pyramid network and a small target layer, and a BiFPN feature pyramid network and a small target layer and a convolution operation attention mechanism module.

[0015] Preferably, the different models in S40 include YOLOv8, YOLOv9, YOLOv10, YOLOv11 and a YOLOv8 network structure model that simultaneously uses a BiFPN feature pyramid network, a small target layer, and a convolution and attention fusion mechanism module.

[0016] Preferably, the S50 attention mechanism comparison experiment includes TA attention mechanism, CBAM attention mechanism, SimAM attention mechanism, ECA attention mechanism and convolution and attention fusion mechanism.

[0017] Compared with the prior art, the wood board edge detection experimental method based on improved YOLOv8 disclosed in the present invention has at least the following beneficial effects:

[0018] Compared with the original YOLOv8 algorithm, the improved algorithm has an improvement of 2.4% and 1.3% in average detection accuracy mAP50 and mAP50:95 respectively, and the detection speed can meet the needs of real-time detection of wood board edge defects. Compared with other detection algorithms in the prior art, it also has an advantage in average detection accuracy, which shows the effectiveness of the improved model. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to make the purpose, technical solution and beneficial effects of the present invention clearer, the present invention provides the following drawings for illustration:

[0020] Figure 1 This is a flowchart of the steps of the experimental method for detecting edge banding of wooden boards based on improved YOLOv8 according to an embodiment of the present invention;

[0021] Figure 2 This is a comparison chart of YOLOv8 model training parameters of a wood board edge banding detection experimental method based on improved YOLOv8 according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] See also Figure 1 , the step flow chart of the experimental method for detecting edge banding of wooden boards based on improved YOLOv8 of the present invention comprises the following steps:

[0024] S10, experimental environment preparation;

[0025] S20, selection of evaluation indicators;

[0026] S30, ablation experiment;

[0027] S40, model comparison experiment;

[0028] S50, attention mechanism comparison experiment.

[0029] S10, experimental environment preparation includes establishing data sets and setting training parameters.

[0030] The dataset includes at least 6,000 images of wood board edge defects. The defect types include: debonding, short tape, residual glue, chipping, glue seam, excessive length, residual tape and dirt. The dataset is divided into training set, validation set and test set in a ratio of 8:1:1. The experimental environment used is shown in Table 1.

[0031] Table 1 Experimental environment

[0032]

[0033] The dataset is divided into training set, validation set and test set in a ratio of 8:1:1. The model training parameter settings are shown in Table 2.

[0034] Table 2 Training parameter settings

[0035]

[0036] The evaluation indicators in S20 include: recall rate R, mean average precision mAP50 and mAP50:95.

[0037] The recall rate R (Recall) indicates the proportion of samples predicted to be positive among samples that are actually positive. The calculation formula of the recall rate is as follows:

[0038]

[0039] Where TP is the number of correctly predicted positive samples, and FN is the number of incorrectly predicted positive samples.

[0040] mAP is one of the commonly used indicators for evaluating target detection models. It takes into account the detection performance of different categories and integrates them into a unified evaluation indicator. The mAP indicators used in this invention are mAP50 and mAP50:95. m represents the mean. AP50 refers to the average detection accuracy when the IoU threshold of the confusion matrix of all categories in the data set is 0.5; AP50:95 refers to the average detection accuracy when the IoU threshold of the confusion matrix of all categories is 0.5:0.95. Among the indicators of target detection, the higher the mAP value, the better the model detection performance. The formula is:

[0041]

[0042] The ablation experiments in S30 include 5 groups, namely, YOLOv8 structure, BiFPN feature pyramid network, small target layer, BiFPN feature pyramid network and small target layer, and BiFPN feature pyramid network and small target layer with convolution and attention fusion mechanism (CAFM) module.

[0043] The experimental results are shown in Table 3.

[0044] Table 3 Ablation experiment based on improved YOLOv8

[0045]

[0046] From the ablation experiment results in Table 3, it can be seen that, taking the mean average precision mAP as an example, the mAP050 value of the YOLOv8n detection algorithm is 60.2%, and the mAP50:95 value is 31.9%. After adding the BiFPN structure alone to model 2, the mAP50 value of the model increased by 0.4%, and the mAP50:95 value increased by 0.9%. After adding the small target detection layer alone to model 3, the mAP50 value of the model was 60.9%, and the mAP50:95 value was 32.5%, which were 0.7% and 0.6% higher than the original algorithm, respectively. After model 4 integrated the BiFPN structure and the small target detection layer, the mAP value of the model reached 62.4% and 33.2%, which was 2.2% and 1.3% higher than the original model. Model 5 adds the CAFM attention mechanism on the basis of model 4. From the results, the mAP value of model 5 reaches 62.6%, which is 0.2% higher than model 4. Through the data analysis of the ablation experiment results, it can be seen that the improved module proposed in the present invention has an improvement effect on model detection.

[0047] The different models in S40 include YOLOv8, YOLOv9, YOLOv10, YOLOv11, and a YOLOv8 network structure model that uses a BiFPN feature pyramid network, a small target layer, and a CAFM attention mechanism. A comparative experiment was conducted on a self-built dataset, and the experimental results are shown in Table 4.

[0048] Table 4 Model comparison experiment

[0049]

[0050] As can be seen from Table 4, the improved YOLOv8 model has better indicators in mAP50 and mAP50:95. Compared with the results of YOLOv9, YOLOv10, and YOLOv11 listed in the table above, the mAP50 of the improved model increased by 1.7%, 4.8%, and 3%, respectively, and the mAP50:95 increased by 0.8%, 2.3%, and 1%, respectively. From the final results, the mAP50 index of the improved model reached 62.6%, and the mAP50:95 rose to 33.2%, which was a significant improvement compared to the original YOLOv8. This shows that the model can identify defects more accurately. At the same time, the FPS of the improved model only dropped by 0.8 to 62.5, which has little impact on real-time detection.

[0051] The mAP results of the model during training can be found in Figure 2 ,As can be seen from the figure, the accuracy curve of the improved model (YOLOv8n-improved) increases more obviously ,in the iterative process compared with the existing YOLOv8n model, indicating that ,the improved model has better convergence ability.

[0052] The S50 attention mechanism comparison experiment includes TA attention mechanism, CBAM (Convolutional Block Attention Module) attention mechanism, SimAM (A Simple, Parameter-Free Attention Module for Convolutional Neural Networks) attention mechanism, ECA (Efficient Channel Attention for Deep Convolutional Neural Networks) attention mechanism and CAFM (Convolutional and Attention Fusion Module) convolution and attention fusion mechanism. The experimental results are shown in Table 5.

[0053] Table 5 Comparative experiments on attention mechanisms

[0054]

[0055] As shown in Table 5, adding CAFM attention significantly improves the detection ability of the model better than other attention mechanisms. Comparing the mAP indicators, after adding TA attention, the mAP indicators of the model reached 61.6% and 32.8%, and the regression rate reached 59.2%. Compared with the basic model, the regression rate increased by 1.9%, but mAP50 and mAP50:95 both decreased. After adding CBAM attention, the mAP indicators of the model were 62.4%, 32.8%, and the regression rate reached 58.9%. Compared with the baseline model, mAP50:95 decreased by 0.4%, and the recall rate increased by 1.6%. After adding ECA attention, the recall rate and mAP50 of the model were 57.4% and 62.1%, respectively. The recall rate increased by 0.1% compared with the baseline model, and the mAP decreased by 0.3%.

[0056] From the experiments, we can see that compared with the TA, CBAM, and SimAM attention mechanisms, after adding CAFM attention, the mAP value and regression rate of the model are in a better position, which proves that the CAFM module performs better in the detection of wood board edge banding defects. Therefore, the CAFM module is added as one of the improvements.

[0057] Through the above experiments, the indicative detection effect of the improved YOLOv8 model was obtained, proving the accuracy advantage and effectiveness of the improved model.

[0058] In addition to the above embodiments, the present invention may also have other implementation modes. Any technical solution formed by equivalent replacement or equivalent transformation is within the protection scope required by the present invention.

[0059] The present invention is described in detail above, but the specific implementation of the present invention is not limited thereto. Those skilled in the art may make various modifications or alterations without departing from the spirit and scope of the claims of the present application.

Claims

1. A wood board edge detection experimental method based on improved YOLOv8, characterized in that: The following steps are involved: S10, experimental environment preparation; S20, selection of evaluation indicators; S30, ablation experiment; S40, model comparison experiment; S50, attention mechanism comparison experiment.

2. The wood board edge detection experimental method based on improved YOLOv8 according to claim 1 is characterized in that: The experimental environment preparation in S10 includes establishing a data set and setting training parameters.

3. The wood board edge detection experimental method based on improved YOLOv8 according to claim 2 is characterized in that: The established data set includes at least 6,000 images of wood board edge banding defects, and the defect types include: debonding, short band, residual glue, chipping, glue seam, excessive length, residual band and dirt. The data set is divided into a training set, a validation set and a test set in a ratio of 8:1:

1.

4. The wood board edge detection experimental method based on improved YOLOv8 according to claim 2 is characterized in that: The training parameters include: initial learning rate 0.01, final learning rate 0.01, number of threads 8, weight decay coefficient 0.0005, momentum 0.937, batch size 16 and number of iterations 250.

5. The wood board edge detection experimental method based on improved YOLOv8 according to claim 1 is characterized in that: The evaluation indicators in S20 include: recall rate R, mean average precision mAP50 and mAP50:

95.

6. The wood board edge detection experimental method based on improved YOLOv8 according to claim 1 is characterized in that: The ablation experiment in the S30 includes 5 groups, namely, a YOLOv8 structure, a BiFPN feature pyramid network, a small target layer, a BiFPN feature pyramid network and a small target layer, and a module that uses a BiFPN feature pyramid network and a small target layer and a convolution operation attention mechanism.

7. The wood board edge detection experimental method based on improved YOLOv8 according to claim 1 is characterized in that: The different models in S40 include YOLOv8, YOLOv9, YOLOv10, YOLOv11 and a YOLOv8 network structure model that simultaneously uses a BiFPN feature pyramid network, a small target layer and a CAFM attention mechanism module.

8. The wood board edge detection experimental method based on improved YOLOv8 according to claim 1 is characterized in that: The S50 attention mechanism comparison experiment includes TA attention mechanism, CBAM attention mechanism, SimAM attention mechanism, ECA attention mechanism and CAFM attention mechanism.

Images