Convolutional Neural Network-Based Identifying Gender of Kiwifruit Flowers in Autonomous Pollination for Future Farming

In recent years, autonomous pollination has become a prevalent topic since it can be an excellent method to mitigate the influence of biological and labour-related variables on the pollination process. This study mainly focuses on the flower detection part of the process (using the kiwi flower as a prototype), which involves flower recognition and gender recognition (distinguishing the stamen and pistil). The present study utilized the YOLOv5 model for object detection. Additionally, three CNN models, namely LeNet, AlexNet, and ResNet, were trained to recognize the gender of the kiwi flowers. The performance of these models was evaluated and compared. Upon careful analysis, it was evident that the LeNet model was most effective in identifying the gender of kiwi flowers, with a test set accuracy of 91%. This result suggests that the trained LeNet model can perform better to accurately recognize the gender of kiwi flowers.