基于多特征光谱的番茄分选算法研究

Study on tomato sorting algorithm based on multiple characteristics spectrum

为了实现番茄成熟度的自动分选,基于番茄可见-近红外波长的多特征光谱,设计一种环境光减除、双信号标记(Dual Signal Marking,DSM)的番茄分选算法。采用均值滤波、环境光减除的方法,有效消除实际干扰对信号的影响。通过研究番茄漫反射的光谱特征,DSM选用520 nm、620 nm、850 nm三种波长作为特征信息。利用TensorFlow分别建立双信号波长520 nm、620 nm的阈值函数φ(x)、γ(x)。将检测信号与所设函数阈值比较,使用标记器获取合格信号占总体信号的比例,从而判断番茄的成熟度。在工厂、农田现场等环境下进行测试。实验结果表明,该算法实现了对番茄成熟度的多档判断,平均识别率大于94. 0%,具有较强的抗干扰能力。

In order to realize the automatic sorting of tomato maturity,according to multiple characteristics spectrum of visible-near-infrared wavelength of tomato,a tomato sorting algorithm based on environmental light subtraction method and dual signal marking method(DSM)was designed.By means of mean filtering and environmental light subtraction,the algorithm can effectively eliminate the influence of actual interference on the signal.By studying the spectral characteristics of tomato diffuse reflection,the wavelengths of 520 nm,620 nm and 850 nm were selected as the DSM characteristics information.By using TensorFlow,the functions of dual signal wavelength 520 nmφ(x)and 620 nmγ(x)were respectively established.The detection signal was compared with threshold value of the function,and the proportion of qualified signal in the total signal was obtained by the marker so as to judge the maturity of tomato.After testing in factory,field,etc.the experimental results show that the algorithm can realize multigrade judgment of tomato maturity.In addition,its average recognition rate is higher than 94.0%and it owns strong anti-interference ability.