Tao Jiang

To address the issues of high breakage rates and substantial labor costs in pigeon egg farming, this study proposes an intelligent pigeon egg recognition and positioning system based on an improved YOLOv12n object detection algorithm and OpenCV barcode recognition technology. Visual sensors installed on feeding machines were used to collect real-time video data of pigeon cages, with images obtained through frame extraction. The images were annotated using LabelImg to construct a pigeon egg detection dataset containing 1500 training images, 215 validation images, and 215 test images. After data augmentation, the dataset was used to train the pigeon egg recognition model. Additionally, customized barcodes were designed according to actual farm conditions and recognized using OpenCV through preprocessing steps including grayscale conversion, filtering, and binarization to extract positional information. Experimental results demonstrate that the proposed YOLOv12n-pg recognition model requires only 4.9 GFLOPS computational load, contains 1.56 M parameters, and has a model size of 3.5 MB, significantly lower than other models in the YOLO-n series. In inference tests, it achieved 99.4% mAP50 and 83.6% mAP50-95. The implementation of a majority voting method in practical testing further reduced the missed detection rate. The system successfully records “cage location—egg count” information as key-value pairs in a database. This system effectively enables automated management of pigeon eggs, improves recognition performance, and demonstrates higher efficiency and accuracy compared to manual operations, thereby establishing a foundation for subsequent research in pigeon egg recognition.

In response to the high breakage rate of pigeon eggs and the significant labor costs associated with egg-producing pigeon farming, this study proposes an improved YOLOv8-PG (real versus fake pigeon egg detection) model based on YOLOv8n. Specifically, the Bottleneck in the C2f module of the YOLOv8n backbone network and neck network are replaced with Fasternet-EMA Block and Fasternet Block, respectively. The Fasternet Block is designed based on PConv (Partial Convolution) to reduce model parameter count and computational load efficiently. Furthermore, the incorporation of the EMA (Efficient Multi-scale Attention) mechanism helps mitigate interference from complex environments on pigeon-egg feature-extraction capabilities. Additionally, Dysample, an ultra-lightweight and effective upsampler, is introduced into the neck network to further enhance performance with lower computational overhead. Finally, the EXPMA (exponential moving average) concept is employed to optimize the SlideLoss and propose the EMASlideLoss classification loss function, addressing the issue of imbalanced data samples and enhancing the model’s robustness. The experimental results showed that the F1-score, mAP50-95, and mAP75 of YOLOv8-PG increased by 0.76%, 1.56%, and 4.45%, respectively, compared with the baseline YOLOv8n model. Moreover, the model’s parameter count and computational load are reduced by 24.69% and 22.89%, respectively. Compared to detection models such as Faster R-CNN, YOLOv5s, YOLOv7, and YOLOv8s, YOLOv8-PG exhibits superior performance. Additionally, the reduction in parameter count and computational load contributes to lowering the model deployment costs and facilitates its implementation on mobile robotic platforms.