UAV T-YOLO-Rice: An Enhanced Tiny Yolo Networks for Rice Leaves Diseases Detection in Paddy Agronomy

The paddy agronomy in the Asia-pacific region has gained a prominent role in connection with the major rice production area in over the decades. The research aims to investigate the aerial computing techniques to improve the sky farming techniques. Recently, the enhancement of unmanned aerial vehicle (UAV) and Internet of Things (IoT) with Deep Learning (DL) in paddy agronomy research has ensured the impact on data availability and predictive analytics. In this research, we focus on Deep Learning (DL) for identifying weeds, regions of crop failure, and crop health in paddy crops. Therefore, a DL architecture suitable for application in aerial computing UAV onboard intelligence is necessary. Furthermore, the DL architecture should be stable and consume as few computational resources as possible, given that it is applied on the UAV's onboard system. This paper proposes to use Tiny YOLO (T-Yolo)V4 as the base detector via following modules: (a) YOLO detection layer is added to the T-YOLO v4 to make the network more capable of detecting small objects. (b) Spatial pyramid pooling (SPP), convolutional block attention module (CBAM), Sand Clock Feature Extraction Module (SCFEM), Ghost modules, and more convolutional layers are added to the network to increase the accuracy of the network. Subsequently, a rice leaf diseases data set which contains the labeled images of rice leaf diseases such as Bacterial leaf blight, Rice blast, and brown spot is obtained. In addition, the image augmentations is applied to produce more samples of the three classes to create our own rice leaf diseases data set. Finally, the enhanced UAV Tiny Yolo Rice (UAV T-yolo-Rice) network has obtained the testing mean average precision (mAP) as
$86 \%$
by training the proposed rice leaves' disease data set. More experimental results reveal that our proposed method outperforms the Rice Leaves' Diseases detection model by using the proposed UAV T-yolo-Rice network set can obtain the highest testing Mean Average Precision (mAP) than all the other models from previous studies. Even the Yolo V7 model produced by darknet cannot have the testing accuracy that is higher than the proposed network.