According to engineering requirements, the image recognition method based on projection features is researched. Based on the actual production line without preprocessing, the image is projected to eight directions directly and image feature compression is extracted. Neural network is adopted to recognize the image and the projection is performed normalized feature extraction. Adaptive resonance ART-1 neural network can remember template image features through learning and training, so neuron I representing template image features is chosen. When the features of the image to be identified are input, neuron I2 is represented by network choice. If neuron I and I2 are same, it is thought that the image to be identified is the same as the template image, and otherwise it is thought that the image to be identified is different from the template image. Recognition error is related with the setting of vigilance parameter ρ.And image recognition is better achieved in the way.