RiceNet: a lightweight instance segmentation network for adhesive rice grains

Crop appearance quality detection based on machine vision has attracted more and more attention in recent years. On the sampling inspection table, because rice grains may touch and adhere to each other, if the collected images are not segmented and preprocessed, it will cause errors in the subsequent evaluation of rice appearance quality. In this paper, an improved adhesive rice grains segmentation network named RiceNet based on Mask R-CNN is proposed. RiceNet uses VoVNetV2 as the backbone network to extract image features. Considering that the research object is the adhesive rice grains image, the target category is single and it is relatively easy to detect, the number of convolution layers and channels of VoVNetV2 are adjusted to carry out the lightweight design of the network. Compared with Mask R-CNN, the accuracy rate and recall rate of RiceNet have increased from 87.2% and 89.7% to 89.5% and 92.6% respectively. Furthermore, RiceNet has few structural parameters which is very suitable for integration into embedded and mobile detection equipment.