Abstract:In view of the low resolution of abrasive grain spectrum images obtained by online visual ferrography,the complex and changeable types of abrasive grains,and the complex background of abrasive grain images,the online intelligent identification of abrasive grains faces challenges.In order to realize the multi-target real-time detection and recognition of abrasive grains in online visual ferrography images,a multi-target recognition method of abrasive grains in online visual ferrography images based on yolov5 was proposed.Taking 6 kinds of wear abrasives,namely normal wear abrasive grains,fatigue wear abrasive grains,sliding wear abrasive grains,spherical abrasive grains,oxidized wear abrasive grains,and cutting wear abrasive grains as the research objects,the segmentation and recognition of abnormal wear particles in complex oil environment based on yolov5 deep neural network model.The results show that the intelligent identification model of abrasive particles based on the yolov5 algorithm can realize the real-time identification of multi-objective and multi-type abrasive particles in complex environment.Its recognition speed and accuracy basically meet the requirements of online monitoring of oil,providing technical support for the industrial application of equipment online image visual ferrography technology.