This work has the ambition generate an algorithm able to clearly identify buried antipersonnel mines from GPR data acquisitions. The algorithm is generated as a combination of a convolutional neural network (CNN) and a symbolic data analysis (SDA) process. The CNN is a powerful tool to automatically detect buried objects with even small metal content; the SDA reduces the probability of false positives, i.e. objects identified as mines, even though they are not and has the great advantage of not requiring a predefined dataset. Experimental campaign, conducted on real terrain, has proven the validity of the presented algorithm.