In the process of QFN chip surface defect detection， the accuracy and efficiency of defect detection can be effectively improved by adding the image segmentation step. In view of the low efficiency of traditional image segmentation and the limitations of low precision and poor stability of image segmentation based on intelligent optimization algorithms， this paper proposed a multi-threshold image segmentation method based on Improved Grey Wolf Optimization （IGWO） algorithm. Firstly， the nonlinear factor in the original GWO algorithm was improved to balance the searching efficiency and mining ability of the algorithm. Secondly， the opposition-based learning was introduced to improve the overall quality of the population， and the sine function and the weight of the head Wolf were introduced to improve the grey wolf updating strategy， so as to enhance the diversity and mining ability of the algorithm. Then， the head wolf approach strategy and population mutation strategy were proposed to update the wolf position， so as to balance the convergence performance and the ability to jump out of the local optimal of the algorithm. Finally， Kapur entropy was used as fitness function to obtain the optimal segmentation threshold. The proposed method was compared with the Grey Wolf Optimization algorithm （GWO）， the Grey Wolf Optimization algorithm based on Disturbance and Somersault Foraging （DSF-GWO）， Levy Flight Trajectory-based Salp Swarm Algorithm （LSSA）， and the image segmentation method of the improved Northern Goshawk algorithm（INGO）in the experiments. The experimental results show that： In terms of segmentation time， the proposed method is about 1/2 that of DSF-GWO and 1/4 that of INGO. In terms of segmentation accuracy and stability， for 30 times of QFN chip defect images segmentation， the average Kapur entropy obtained by the proposed method is the largest， and the standard deviation is the smallest. Therefore， the proposed method can realize multi-threshold segmentation of QFN images with high accuracy， high stability and high efficiency.