Abstract: Identification of screen breakage in a hammer millis relying mainly on the manual feed sampling currently in the Chinese feed industry. There are four disadvantages in the manual selection for the feed via artificial senses: a)Human errors and failings easily occurred in the mechanical detection of an online system; b)Low efficiency is difficult to meet the requirements of automatic production in modern agriculture; c)High misdetection rate can result in the unqualified granularity of product; d)The procedure is costly, time-consuming and labor-intensive, sometimes involving hazardous chemicals. In this study, an online automatic identification system mounted on a novel sampling mechanism was established for the rapid and accurate detection of screen breakage in a hammer mill using imaging technology and machine vision. Three key parts were designed in the device, including sampling mechanism, screening mechanism, and image acquisition mechanism. Specifically, the sampling mechanism mainly consisted of a chute feeder and automatic sampling components. The screening mechanism was composed of a vibrating feeding device, screen components, and feeder base. The platform of image acquisition included camera abscura, cloud platforms, and complex optical assembly, such as camera, lens, as well as an annular LED light source. The sampling mechanism was selected to automatically select feed samples in real time. The screening mechanism was applied to screen and separate the obtained samples, thereby transfer the samples to the next mechanism. The platform of image acquisition was then applied to collect the image of particles in the samples. A Siemens S7-200 PLC programmable controller was applied to realize automatic control in the whole system, including sampling, sample screening, vibrating, and image collecting. Python language was used for the image processing and programming, graying, denoising, and binarization that contained in image preprocessing, where the contour of large particles and the minimum circumscribed rectangle were determined during the program. The equivalent diameter of the projected circle and the minimum area of circumscribed rectangular for the screened large particles were taken as the characteristic parameters and the threshold, K-nearest neighbor and radial basis function support vector machine were utilized to establish the identification model of screen breakage. The accuracy rates of the threshold model were 98% and 93% for the diameters of 1.0 and 2.0 mm in the screen identification. The recognition model established by the threshold was not stable to deal with the experimental data, due to the different distribution of dataset. The accuracy rates of the KNN(K-Nearest Neighbor, KNN) model were 94% and 96% for the diameters of 1.0 and 2.0 mm in the screen identification, when k was 3, and the input parameters of the nearest neighbor KNN model were set as the average diameter of the projected circle and the average minimum of the circumscribed rectangular area. When the input parameters of the radial basis kernel function support vector machine (RBFSVM) model were set as the average diameter of the projected circle and the average minimum of the circumscribed rectangular area, and the optimal penalty coefficient and gamma parameters (C, g) were (2-5, 2-7) and (2, 2-1), respectively, the recognition accuracies of RBFSVM model were 89% and 91% for the diameters of 1.0 and 2.0 mm in the screen identification. As such, the KNN recognition model was determined as the recognition of the system, in case there was no significant difference between the running speed of KNN and RBFSVM models. The newly developed device can provide promising technical support to the online automatic identification of screen breakage in a hammer mill.