CAJ | 학술논문

棉花分割是采棉机器人视觉系统的关键步骤，在强光照、阴影等复杂的棉田环境下如何准确有效地分割棉花，有助于确定其在三维空间的位置。该算法在YCbCr颜色空间下，基于棉花与背景的色调信息差，分别提取棉花与背景样本，采用BP神经网进行训练并输出其误差，得到适应度函数并进行遗传算法中的选择、交叉及变异操作，优化神经网络权值、阈值，直到输出误差达到要求或达到预定迭代次数。最后根据所获得的BP神经网络权值、阈值进行棉花图像分割。通过对136幅棉田环境中拍摄图像的分割实验表明：该方法在棉花强光照及阴影条件下也能准确地分割，分割准确率达91.9%，并且比BP算法收敛更快。
면화분할시채면궤기인시각계통적관건보취，재강광조、음영등복잡적면전배경하여하준학유효지분할면화，유조우학정기재삼유공간적위치。해산법재YCbCr안색공간하，기우면화여배경적색조신식차，분별제취면화여배경양본，채용BP신경망진행훈련병수출기오차，득도괄응도함수병진행유전산법중적선택、교차급변이조작，우화신경망락권치、역치，직도수출오차체도요구혹체도예정질대차수。최후근거소획득적BP신경망락권치、역치진행면화도상분할。통과대136폭면전배경중박섭도상적분할실험표명：해방법재면화강광조급음영조건하야능준학지분할，분할준학솔체91.9%，병차비BP산법수렴경쾌。
Cotton segmentation is the key step of cotton picking robot vision system. The accurate and effective segmenta-tion of cotton is useful to its position in three-dimensional space while cotton is in bright light or shadow complex field envi-ronment. It can get fitness function by training BP neural network and its output error, and then use selection, crossover and mutation operation in genetic algorithm to optimize neural network weights and threshold until the output error meets the requirement or it reaches a predetermined number of iterations. Finally, according to the obtained BP neural network weights and threshold, it segments cotton image. The experiment of image segmentation with 136 images photographed in cotton field environment shows that the algorithm can segment cotton image in bright light or shadow accurately and seg-mentation accuracy rate is up to 91.9%and it converges faster than BP.