A wheat seedling counting method based on two-stage convolutional neural network
As one of the world’s major food crops, accurate counting of the number of wheat seedlings is of great significance for subsequent yield prediction. In response to the problems in existing manual counting methods, such as time consuming, labor-intensive, and prone to subjective errors, we proposed a...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Smart Agricultural Technology |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2772375525005635 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | As one of the world’s major food crops, accurate counting of the number of wheat seedlings is of great significance for subsequent yield prediction. In response to the problems in existing manual counting methods, such as time consuming, labor-intensive, and prone to subjective errors, we proposed a new method for automatically counting wheat seedlings based on laser-labeled Convolutional Neural Network (CNN). Firstly, after labelling wheat seedlings with laser, the image data was collected using a self-developed data acquisition device. In order to avoid the influence of laser points shining on the background on the counting results, a two-stage segmentation method was designed for data processing. In the first stage, the SCNet model was applied for wheat seedling segmentation, the ResNet50 model was chosen as the backbone network, and the Convolutional Block Attention Module (CBAM) was introduced for feature enhancement. Then, the Swin Transformer Block self-attention mechanism was used to fuse the enhanced features with features in the basic module. In the second stage, the SS-DeepLabV3+ model was applied for laser point segmentation, the MobileNetV2 model was chosen as the encoder, and the Squeeze-and-Excitation (SE) attention mechanism and self-attention mechanism were introduced to enhance the model’s feature representation. Finally, an automated counting application for wheat seedlings was developed, which works by counting the number of laser points. The proposed two-stage algorithm showed significant advantages in terms of evaluation metrics including the mean Pixel Accuracy (mPA), Recall, and mean Intersection to Union (mIoU). This study provides an efficient and reliable method for the automated counting of wheat seedlings, which is conducive to promoting the development of precision agriculture. |
|---|---|
| ISSN: | 2772-3755 |