A Fragmented Target Recognition System Based on Zero-Shot Learning

In recent years, target recognition and detection methods based on deep learning have shown great application prospects in many fields, such as smart house, driverless technology, product detection and military equipment, etc. However, in some extreme application scenarios, such as the emergency rescue, the target is inevitably fragmented due to the impact of explosion and many other factors, which leads the lack of effective feature information in the target image and affects the accuracy of target recognition and classification. In order to solve this problem, this paper proposes a new method to recognize fragmented targets based on zero-shot learning. This method solves the problem of target recognition under the condition of zero samples by introducing some high-level attributes. For verifying the effectiveness of this method, this paper takes five kinds of ingredients after cutting in daily life: cucumber, potato, tomato, eggplant, and bamboo as an example to illustrate and verify the whole process of the feature extraction, attribute recognition and the target classification in this method. The experiment results show that the highest recognition accuracy for the ingredients after processing in this fragmented recognition system is 76%. In addition, this paper also develops and verifies the real-time recognition of this system on the embedded platform PYNQ.