Accurate Strawberry Plant Detection System Based on Low-altitude Remote Sensing and Deep Learning Technologies

Aiming at solving the problem of real-time monitoring strawberry plant growth in fields, a new scheme based on low-altitude remote sensing technology combined with deep learning is proposed to identify the strawberry plants. The goal is to establish an accurate, fast and reliable strawberry plant counting system, which is a key part to realize the fine management of strawberry planting in fields. To accurately recognize the strawberry plant objects in the complex growing environment, the outer contour characteristics of strawberry plant canopy in the low-altitude remote sensing images in the visible spectrum band is proposed to define the criterion of discriminating the strawberry plants. The state-of-the-art detection framework Faster R-CNN with different deep hierarchical structures is used to train the strawberry plant detection models. The results show that the average precision of the Faster R-CNN with the ResNet-50 backbone reaches 0.9888, which is 2.51% higher than the VGG-16-based model. The presented model can achieve the same accuracy as human beings. In the future, the system is expected to replace the artificial task of identifying the strawberry plants to cut the cost of strawberry production.