Abstract: To improve accuracy and speed of tomato quality recognition, a recognition method based on an improved YOLOv5 was proposed．Method was based on YOLOv5 as a recognition framework．It replaced spatial convolution module of CSPDarknet53 network in YOLOv5 with Bottleneck Transformer（BoT3）module to improve network computing speed．Attention mechanism to enhance feature expression ability of YOLOv5 was introduced, and replacing CIoU loss function of YOLOv5 with SIoU．Finally, improved YOLOv5 was used to identify tomato quality．Results showed that proposed method achieved an average recognition accuracy, recall, and accuracy of 90.11%, 95.21% and 95.10% for red and green tomatoes' quality, with an average detection speed of 129 frames per second．Compared with SSD, CNN and VGG, improved YOLOv5 had obvious recognition advantages．It was proved that improved model could improve recognition accuracy and speed of crop quality of such as tomatoes．