Mechanical damage of tea plant is a serious problem in tea production. This work employed metal oxide semiconductor (MOS) gas sensors and gas chromatography-mass spectrometer (GC-MS), as an auxiliary technique, to detect tea plants with different types of mechanical damage in different severities. Various algorithms were applied. The results showed the uniformity of the results of gas sensors and GC-MS. While, it was hard for gas sensors to discriminate among tea plants with different types of mechanical damage. However, the feasibility of gas sensors for predicting the damage severity in different damaged types based on gas sensors was proven, which was more meaningful. Finally, multi-layer perceptron neural networks (MLPNN) was employed and the results showed that the correct discrimination accuracy rate for damage severity was 99.07% for the training set and 95.83% for the testing set, which indicated that MLPNN was an excellent algorithm for damage severity determination. This study provided a new technique for mechanical damage of tea plant detection and was very meaningful for tea plant protection.
Read full abstract