Recognition of Grape Species with Small Samples Based on Attention Mechanism

Abstract

Take Turpan as an example, in recent years, the grape industry in Turpan has become one of the most important pillar industries for rural economic development and farmers’ income increase in Turpan. However, there are still many problems in the development of grape industry in Turpan area. Due to many and complicated grape varieties, it is difficult to identify them, and current identification efficiency is far from enough. Moreover, there is no large-scale open data set in grape recognition at present, and each image itself taken for the grape has much noise effect, which leads to low recognition accuracy. In this paper, a small sample of grape variety recognition method based on attention mechanism is used to fine-tune the CNN model and process the dataset image differently, and is compared with the traditional method. The experiment results show that the method can recognize different grape image types accurately with an accuracy of 93.72% under the condition of small sample. By the method, we can not only improve the efficiency of intelligent recognition, but also reduce the manpower cost, and thus realize the intelligent recognition of grape types.