The work aims to carry out intelligent elimination, identification and early warning of the transverse filter tip produced during the operation of a transmitter to improve the efficiency of the production based on image recognition and structure improvement. The lateral transport images of filter rods were collected, and the abnormal images were tracked, from which the feature points of abnormal filter rods were extracted, and the warning information was generated based on the images. At the same time, the structure optimization of the backplane of the transmission track of the filter rods transmitter was carried out to eliminate most of the transverse filter tips. In the improved simulation prototype, the rejection rate of the transverse filter tip was as high as 77% when the transverse filter tip was transported to the conveying track backplane, and the other abnormal transverse filter tip was immediately warned. The blockage probability of the nozzle transmitter pipeline was reduced by about 60%, and the operating efficiency of the equipment was significantly improved. Therefore, image recognition and structure improvement can effectively identify and eliminate the transverse filter tip generated during the operation of the transmitter.