Detection Method And Application Of Diseased Banana Trees Based On Multi-source Data Fusion

Banana is one of the most important economic crops in the world and one of the most important fruits in the world.Its planting area and output are very extensive all over the world.Banana trees are susceptible to disease.For a long time,the disease identification of banana trees has relied on human identification or automatic identification of a single data.To solve this problem,this paper proposes a multisource data fusion monitoring method for diseased banana trees.The disease is detected in many aspects,and the specific research contents and conclusions are as follows:1.Aiming at the problem of a single data source,this paper collects banana disease image data from high altitude and ground through different equipment,and establishes a multi-source disease banana tree data set.For the problem of less diseased banana tree data,the original disease banana tree data is analyzed.enhanced.By building different target detection network models,and then training the diseased banana tree dataset through migration learning,comparing the training results,and finally selecting YOLOv5 as the basic network model for diseased banana tree recognition.2.An improved diseased banana tree identification method based on the YOLOv5 model is proposed,and the attention mechanism is introduced,and the CBL module of the YOLOv5 network is replaced with GhostNet,which improves the network performance and reduces the calculation amount of the network model.The recognition performance of the improved model was trained and tested,and the results met expectations.Based on the generated model,a method for monitoring diseased banana trees with multi-source data was proposed.3.With the promotion of smart agriculture and the development of embedded technology,in order to be more efficient in application and more convenient for users to use,a diseased banana tree detection platform has been designed,and highaltitude orthophoto recognition and ground silhouette recognition have been realized for multi-source data.fruit recognition and camera real-time recognition and other functions.