In the process of automatic targeted spray in forest region at present, the accuracy and efficiency of point cloud data are all low when the tree trunks grow intensively, aimed at which the optimized K-means clustering algorithm was put forward, and data acquisition method was based on 2D laser detection. In view of the relevant data needed to be filtered before clustering analysis for trunk scanning spots, application of window filtering algorithm was put forward. The edge of trunk which generated mixed pixels was selected, and then the mixed pixels deriving from three adjacent scans and the scanning spots deriving from two scanning angles near the mixed pixel were extracted, finally, the maximum threshold filtering processing for the neighbor spots was done. Through 50 times of extractions and analyses of test data, only two mixed pixels were not filtered, which indicated that the filtering rate of mixed noises was high. Aimed at the defects of cluster number and initial cluster centers for Kmeans algorithm needed to be predetermined, the method of slope variation used to determine the clustering number was firstly proposed. Five groups of trunks were respectively measured for 100 times at five different distances in the test, and results showed that the number of error measurements was only three times, which could be removed by artificial way at the early stage of the test, and it indicated that the slope variation algorithm was reasonable and effective. The performance of Huffman tree method, which was used to determine the clustering centers for the trunk scanning spots, was analyzed in another test, and K-means clustering was carried out by using random sampling method and Huffman tree method under three trunk distribution types. The average correct rate of former was only 76.4%, while that of the latter was 95.5%. Meanwhile, iterations and time-consuming using the two above-mentioned algorithms at type I distribution were analyzed, and the average number of iterations of random sampling method was obviously higher than that of Huffman tree method at five different distances, but the average time-consuming of Huffman tree method was higher than that of random sampling method. The variation range of former was 120~220ms and it was 50~85ms for the latter, which were all in acceptable ranges on forest surveying and mapping. Experiments proved that the determining methods for clustering number based on the slope variation algorithm and clustering centers based on Huffman tree method were effective algorithms for the clustering of trunk scanning spots in forest region during using K-means algorithm, which could be applied to tree trunk detection for actual forest region.