基于深度学习的蚕茧种类实时检测系统设计

Design of real-time detection system for silkworm cocoon species based on deep learning

针对目前人工选茧效率低、自动化辨识技术尚未被有效利用等问题,设计了一种基于深度学习的蚕茧种类实时检测系统。该系统主要由检测装置、蚕茧图像识别模块、气泵吹气控制模块及蚕茧种类检测管理模块组成。蚕茧分批进入粗选机构中,该机构以振动分离的方式快速分离部分下茧;剩余蚕茧通过单粒化机构进行有序排列;通过精选机构拍摄蚕茧完整表面图像,以深度学习模型BL-YOLOv3为依托,对黄斑茧、上车茧及烂茧进行有效识别;随后控制气泵模块对已识别的蚕茧进行吹气分离;最后将蚕茧图像、检测结果等信息存储于蚕茧种类检测管理模块中,可实时显示蚕茧种类识别结果。测试试验表明,该系统能够完成6种蚕茧的实时检测,较传统人工选茧模式更为方便快捷。

Aiming at the low efficiency of manual cocoon selection and the fact that automatic identification has not been effectively used,a real-time detection system of cocoon species based on deep learning is designed.The system is mainly composed of a detection device,a cocoon image recognition module,an air pump blowing control module,and a cocoon species detection management module.The cocoons enter the rough selection mechanism in batches,which quickly separate part of the cocoons by vibration separation;the remaining cocoons are arranged in an orderly manner through the single granulation mechanism;the complete surface image of the cocoons is captured by the selection mechanism,and relying on the deep learning model-BL-YOLOv3 model,it could effectively identify macular cocoons,reelable cocoons and rotten cocoons;then control the air pump module to blow and separate the identified cocoons;finally the cocoon image,detection results and other information are stored in the cocoon species detection management module,it could display the result of cocoon type recognition in real time.Test experiments show that the system could complete the detection experiment of six kinds of cocoons,and realize the real-time detection process,which is more convenient and quicker than the traditional manual cocoon selection mode.