Camellia Oleifera Fruit Detection Algorithm in Natural Environment Based on Lightweight Convolutional Neural Network

At present, Camellia oleifera fruit harvesting relies on manual labor with low efficiency, while mechanized could result in bud damage because flowering and fruiting are synchronized. As a prerequisite, rapid detection identification urgently needed for high accuracy efficiency simple models to realize selective intelligent harvesting. In this paper, lightweight algorithm YOLOv5s-Camellia based YOLOv5s is proposed. First, the network unit of ShuffleNetv2 was used reconstruct backbone network, thereby number computations parameters model reduced increase running speed saving computational costs. Second, mitigate impact improvement accuracy, three efficient channel attention (ECA) modules were introduced into enhance network’s features, Concat operation neck replaced by Add fewer parameters, which amount information under features maintaining same channels. Third, Gaussian Error Linear Units (GELU) activation function improve nonlinear characterization ability network. addition, locate objects natural environment, penalty index redefined optimize bounding box loss function, can convergence regression accuracy. Furthermore, final experimental results showed that possesses 98.8% 5.5 G FLOPs computation, 6.3 MB size, reached 60.98 frame/s. Compared original algorithm, calculation amount, 65.18%, 56.55%, 57.59%, respectively. The provide technical reference development fruit-harvesting robot.