A Research of Neural Network Optimization Technology for Apple Freshness Recognition Based on Gas Sensor Array.

In the process of growth, apples' ripening, storage, transportation, and processing, the appearance and internal physiological characteristics will have some changes because of the effect of time and physical attributes. A series of problems, such as the false ripeness and putrefaction of fruit, will bring huge economic losses to fruit vendors and harm to consumers. In this paper, an odor recognition system has been designed for the fast evaluation of the freshness characteristics of apples, which is based on the freshness characteristics of Fuji apple. A series of apple-air mixture with equivalent model was established by studying the change of gas concentration during the growth and storage of apples. The continuous projection algorithm (Successive Projections Algorithm, SPA) is used to optimize the sensor array to solve the problems of collinearity and overlap and also to eliminate the abnormal and redundant sensors. ZigBee wireless sensor network is adopted to send data to host computer, and BP (Error Backpropagation) neural network algorithm optimized by SFLA (shuffled complex evolution, SCE + Particle Swarm Optimization, PSO) algorithm is used to recognize gas data, which greatly improves the training speed and precision of neural network. The experimental results show that the detection accuracy of the Fuji apples freshness is 98.67% and can quickly and comprehensively identify the freshness of apples. [ABSTRACT FROM AUTHOR]