Research on the power consumption of the sugarcane stalk cutting system is beneficial to improve the cutting performance of the sugarcane harvester and the utilization rate of the engine power. Therefore, in order to obtain the change of the cutting force and cutting power of the sugarcane cutting device in the working process, the article adopts the unit combination method and combined with ANSYS/Ls-Dyna to numerically simulate the cutting process of the harvester cutting device. The cutting blade linear speed, the cutting blade angle and the cutting edge angle are used as experimental factors, and the cutting power consumption and cutting force are used as the experimental indicators to conduct a singlefactor experimental analysis. The parameter range of the cutting experimental factors is determined and the simulation experiment design is carried out at the same time. The minimum power consumption is the optimization goal, and the best cutting conditions are that the speed of the contact point of the cutting knife and the cane stem is 38.8 m/s, the cutter head angle is 11.66°, and the blade angle is 25°. The cutting device of sugarcane harvester consumes the least power when cutting sugarcane stems, and its minimum power is 0.80 kW.

In order to obtain high accuracy of plant disease classification with only a few training samples, a few-shot learning model was used as the disease classifier, and five kinds of shallow networks, Conv4, Conv6, ResNet10, ResNet18, and ResNet34, were used as feature extraction networks under the framework of three typical few-shot learning algorithms, including MatchingNet, ProtoNet, and RelationNet. Their performances were compared on the plant disease data set of PlantVillage. Under the condition of 5way、1shot, the average accuracies of MatchingNet, ProtoNet, and RelationNet were 72.29%, 72.43%, and 69.45%, respectively. ProtoNet+ResNet34 was the optimal combinational mode, and the accuracy reached 77.60%. Under the condition of 5way、5shot, the average accuracies of MatchingNet, ProtoNet, and RelationNet were 87.11%, 87.50%, and 82.92%, respectively. The accuracies were significantly improved compared to that of the 1shot condition. ProtoNet+ResNet34 was still the optimal one with an accuracy of 89.66%. The above test results show that by optimizing the combination of a few-shot learning framework and feature extraction network, the recognition model can achieve good effects for diseases with a small number of samples.