Deep Learning Green Citrus Detection and Counting in Orchards Based on YOLOv5-CS and AI Edge System Papers

Green Citrus Detection and Counting in Orchards Based on YOLOv5-CS and AI Edge System

author

School of Electronic Engineering, South China Agricultural University

achievement

Realized the detection and counting of citrus, the algorithm is based on YOLOv5 and DeepSORT (deployed on the edge device)

Innovation

1. Data augmentation is used , including horizontal and vertical mirroring, translation, blurring, rotation 270 degrees, adding noise, etc.
2. Add the CBAM attention mechanism to the convolutional layer after the Focus layer .
3. On the basis of the original three target detection layers, a small target detection layer is added (the original feature map output size is 52x52, 26x26 and 13x13, and now the size of 104x104 is added).
4. Change GIoU loss function to CIoU loss function .
5. The algorithm is deployed in the edge device .
6. The adjustment of the learning rate adopts the cosine annealing algorithm (included in YOLOv5).
7. In tracking and counting, the concept of " virtual line " or " virtual area " is introduced (similar to the idea of ​​the laboratory students), that is, the number of citrus will increase if the citrus crosses a set line.