In order to identify and locate the maize rhizomes accurately, a maize rhizome detection network based on the migration learning method was established. The function of human eye recognition to identify and locate the rhizomes of the corn from a complex field environment was simulated, which achieved the function of crawler heat fog machine walking along the corn line. Field image of corn was collected by crawler self-propelled hot fogging machine, construction of a precise identification and location model of corn rhizome based on convolutional neural network, and the “DOG Pyramid” algorithm was used to extract maize rhizome as the target from the images, which constituted the training sample database. Through training network, the single maize rhizome was precisely identified firstly, and then were accurately identified and located in the environment of corn crop. The path tracking was obtained by east square fitting algorithm based on the identified maize rhizome location, and the sliding mode track tracking algorithm was used to control the double differential drive motor of the caterpillar chassis to realize the path tracking. The test result showed that the corn root recognition method can identify and locate the maize rhizomes more accurately, the correct rate of identification and location of corn rhizome reached 91.4%, but the traditional image processing method can only reach 67.3%. It can be seen that the method of identifying maize rhizomes proposed had better positioning accuracy, which can better plan the corn field path accurately. The research results provided the key technical support for the crawler self-propelled hot fogging machine self walking along the intercropping of corn.