Abstract: In order to identify extractable uranium ore and enhance resource utilization efficiency,utilizing X-ray transmission technology in conjunction with the semi-supervised learning algorithm - ITSVM,to achieve intelligent uranium ore sorting. At the same time,in order to further optimize the performance of the model,the light and dark correction method is introduced to solve the problem of image noise. This method maps each pixel in the noisy image through normalization processing, so as to improve the image quality. Through the improved straight concave detection and cutting algorithm and slicing method, the difficult problem of multi-objective classification task of support vector machine is overcome. The algorithm detects the position and distance of pixels relative to the straight line,uses constraints to judge the concave point,and obtains the corresponding cutting line by using the minimum distance cutting method. Then the multi-target detection problem is transformed into several independent single-target detection problems by slicing method. By synthesizing these two optimization methods,the ITSVM uranium ore sorting model is finally established. The model was trained and tested by 2 000 uranium ore images collected by Xray projection technology,and the results were compared with SVM and TSVM models. The test results show that the accuracy of the model in detecting uranium ore is improved by 2. 9 percentage points after light and dark correction;The ITSVM model has the function of multi-target detection by using improved straight concave detection and cutting algorithm and slicing method, and the accuracy of the model for multi-target uranium mine image detection reaches 95. 7%. On the test set,the accuracy of ITSVM model to detect uranium ore reaches 97. 3%. Compared with SVM and TSVM,ITSVM has greater advantages in the accuracy of uranium ore detection and continuous optimization model. The experimental results verify the feasibility of ITSMV model in the field of uranium ore sorting.