Blueberries are one of the more economically rewarding fruits for fruit growers. Identifying blueberry fruit at different stages of maturity is economically important and can aid fruit growers in planning pesticide applications, estimating yields, and efficiently conducting harvesting operations, among other benefits.
Visual methods for identifying the different ripening stages of fruits are increasingly receiving widespread attention. However, due to the complex natural environment and the serious shading caused by the growth characteristics of blueberries, the accuracy and efficiency of blueberry detection are reduced to varying degrees. To address the above problems, in the study presented herein, the researchers constructed an improved YOLOv9c detection model to accurately detect and identify blueberry fruits at different ripening stages. The size of the network was reduced by introducing the SCConv convolution module, and the detection accuracy of the network in complex and occluded environments was improved by introducing the SE attention module and the MDPIoU loss function.
Compared to the original model, the mAP0.5 and mAP0.5:0.95 of the improved YOLOv9c network improved by 0.7% and 0.8%, respectively. The model size was reduced by 3.42 MB, the number of model parameters was reduced by 1.847 M, and the detection time of a single image was reduced by 4.5 ms. The overall performance of the detection model was effectively improved to provide a valuable reference for accurate detection and localization techniques for agricultural picking robots.
Feng, W.; Liu, M.; Sun, Y.; Wang, S.; Wang, J. The Use of a Blueberry Ripeness Detection Model in Dense Occlusion Scenarios Based on the Improved YOLOv9. Agronomy 2024, 14, 1860. https://doi.org/10.3390/agronomy14081860