Research on Improved Mask R-CNN Recognition Technology in Complex Silkworm Background

This paper proposes an improved Mask-RCNN_ResNeXt101 model to address the issues of background noise and blurry silkworm body contours in existing silkworm body recognition technologies. This method enhances feature extraction capability by using ResNeXt101 in the backbone network, helping the model to more clearly distinguish the contour boundaries of silkworms in complex backgrounds. With the advancement of this technique, the goal is to address the issues associated with background interference and indistinct outlines of silkworm bodies that are encountered in current silkworm recognition methods. The experimental results show that using ResNeXt101 as the backbone network, compared to the basic Mask-RCNN_ResNet50 model, the AP value increased by 3.7%, the recall rate increased by 7.0%, and the detection accuracy improved by 3.6%. The experimental results show that the use of ResNeXt101 in the backbone network exhibits significant advantages in background noise and silkworm body contour blurring tasks, especially in scenes with high silkworm body background interference, where the Mask-RCNN_ResNeXt101 model can accurately identify silkworms. The application of this technology not only improves the accuracy of silkworm body recognition, but also provides new ideas for image recognition tasks in agriculture, promoting the intelligent and modern development of sericulture .