With the emerging of compressed sensing (CS), it is possible to overcome the storage and transmission difficulty of the mass data sampled by traditional methods. It also provides a new way for machine vision applied to apple image sampling. However, the major shortcoming of the reconstruction algorithms for CS signals is the expensive computing time, which limits its applications to the occasions requiring fast processing. Aiming at this problem, two dimensional orthogonal matching pursuit algorithm with parallel computing is proposed for apple image reconstruction. The parallelism of the algorithm is analyzed and the parallel algorithm using CUDA technology on GPU is designed in order to achieve a fast reconstruction algorithm. Experimental results show that the parallel algorithm improves the recovery efficiency by 16 to 35 times and the apple image can be recovered in several seconds. This method provides a new technical support to remote monitoring in real time for apple garden. It can be used in the fast apple quality detection based on image as well.