Paddy and Water Environment - In order to investigate the radiation utilization efficiency and soil temperature with different irrigation methods in cold region and black glebe, northeast, China,... 相似文献
To reduce the consumption of agricultural irrigation water, which is the most part of the total water consumption, choosing the most efficient irrigation method is the best way to save agriculture water. Different from the former studies, a comprehensive evaluation method named Irrigation Method Deep Evaluation Model (IMDEM) was proposed to evaluate irrigation methods of farmland. IMDEM consists of two parts: indicator screening and Deep Evaluate Model. In the first stage, to screen the preliminary indicators and select final indicators which were used for evaluating different irrigation methods, a method named GApriori consisting of generative adversarial network (GAN) and Apriori algorithm was presented. In the second, to choose the most efficient irrigation method, Deep Evaluate Model, which includes GAN and convolutional neural network (CNN). From 2017 to 2019, field experiments were carried out in the semi-humid area in Heilongjiang Province in China. Irrigation methods during the experiment period were chosen control irrigation, wet irrigation, and flood irrigation. And 14 preliminary indicators were summarized in yield and yield components, agronomic characters, photosynthetic characteristics, and resource use efficiency. IMDEM results show that accuracy, macro-average precision rate, macro-average recall rate, and macro-average F1 value were 79%, 81%, 81%, and 79%, respectively. Additionally, after evaluated by IMDEM, control irrigation is the most suitable irrigation method in the semi-humid area of Heilongjiang Province in China. Different from former studies, IMDEM has three contributions: (1) Due to data acquisition difficulty and less data volume in the field experiment, former studies have used many replicate experiments to obtain data. Therefore, the IMDEM proposed in this paper was used GAN to generate a large amount of real-fake data for training the evaluation model. (2) Different from the traditional indicator screening methods that calculate the contribution of each indicator. The IMDEM screening indicators by counting the frequency and association relationship of indicators appearing in each sample. (3) Different from the traditional evaluation method of farmland irrigation methods that only compares rice growth period data. This paper proposed the IMDEM for comprehensive evaluating irrigation methods. And IMDEM has high accuracy for the level prediction of generated data.