Multi-task Tomato Fruit and Bunch Maturity Detection Approach Based on Improved YOLOv7

Abstract

Cherry tomatoes are well loved by people all over the world as a fruit that can be eaten fresh. The detection and ripeness grading of both individual tomato fruits and fruit clusters are necessary conditions for inspection robots to perform inspections. To achieve a three-task detection of tomato fruit and bunch detection, fruit ripeness and bunch maturity grading, a multi-task detection network based on YOLOv7 was proposed in this paper. We first constructed a dataset of cherry tomatoes satisfying the multitasking requirements. Next, the YOLOv7 backbone network was selected and two additional decoders were added to the Head section to simultaneously implement tomato bunch identification, fruit ripeness and bunch ripeness detection. Finally, the multi-task loss function was designed and SIoU was substituted for CIoU to improve the accuracy of model recognition. The experimental results showed that the proposed model had good detection performance and maturity grading without significant increase in memory overhead and network depth. With an IoU threshold of 0.5, the average detection accuracy of cherry tomato bunches and fruit ripeness reached 86.9%, and average inference time was 4.9ms. In the future, it is expected to play an important role in the field of automated inspection.