One of the modern and potentially highly beneficial trends in precision agriculture is application of mobile robots. These robots provide capabilities like planting, watering, fertilizing, harvesting crops, thus saving time and resources. One of the major challenges to be addressed to the mentioned capabilities is the ability to recognize crops, weeds, soil etc. As distinguishing features of crops and their environment have wide variety, the recognition system should have the ability to adapt and learn thereby ensuring recognition of previously unseen scenes. The current state of the art of image recognition methodologies mostly focuses on various types of artificial neural networks that usually are implemented as software imitation. Benefits for using imitation are greater adaptability and easier modification, but the drawback is the required computing power. In mobile robots these resources usually are of great value due to the limited space and power source capacity. As an alternative to imitation we propose to use artificial hardware neural networks. The objective of the presented research is to determine artificial hardware neural network capabilities in image recognition and develop an image recognition system for mobile robots. In this paper existing solutions of artificial hardware neural network usage, solution structure, restrictions and possibilities are examined. Using the only hardware neuron controller available on the market CM1K, its supporting hardware and software are developed and presented for the image recognition sensor. Image recognition capabilities are tested by training neural networks using photos of objects of different colour and shape. In conclusion the acquired recognition results and CM1K capabilities of image recognition are analysed and discussed.