CAJ | 학술논문

针对和声搜索算法存在早熟、收敛停滞等问题，提出了一种基于混合蛙跳细菌觅食的和声搜索算法（harmony search algorithm based on shuffled frog leaping and bacterial foraging，SFLBF-HSA）。引入混合蛙跳算法全局搜索及细菌觅食优化算法群聚吸引、排斥信号等思想，对和声音调搜索机制进行了改进。首先，提出和声音调学习策略（即吸引信号），利用全局最优和声个体对最差和声个体进行正方向的差异扰动，保持搜索朝向最优个体；其次，提出和声音调调节策略（即排斥信号），利用全局最优和声个体对最差和声个体进行反方向的差异扰动，保持搜索远离局部最优个体并朝向其他优秀个体。在两种策略搜索中，同时利用全局共享因子的非线性动态特点抑制搜索的随机性。Benchmark函数对比实验结果表明，改进后算法在单峰值和多峰值函数寻优问题上收敛速度和精度均有显著提高。将SFLBF-HSA应用于作物籽粒图像分割，提高了分割效果，对颗粒较大，似圆形状作物籽粒图像分割后，识出率和识别成功率有显著提高。
침대화성수색산법존재조숙、수렴정체등문제，제출료일충기우혼합와도세균멱식적화성수색산법（harmony search algorithm based on shuffled frog leaping and bacterial foraging，SFLBF-HSA）。인입혼합와도산법전국수색급세균멱식우화산법군취흡인、배척신호등사상，대화성음조수색궤제진행료개진。수선，제출화성음조학습책략（즉흡인신호），이용전국최우화성개체대최차화성개체진행정방향적차이우동，보지수색조향최우개체；기차，제출화성음조조절책략（즉배척신호），이용전국최우화성개체대최차화성개체진행반방향적차이우동，보지수색원리국부최우개체병조향기타우수개체。재량충책략수색중，동시이용전국공향인자적비선성동태특점억제수색적수궤성。Benchmark함수대비실험결과표명，개진후산법재단봉치화다봉치함수심우문제상수렴속도화정도균유현저제고。장SFLBF-HSA응용우작물자립도상분할，제고료분할효과，대과립교대，사원형상작물자립도상분할후，식출솔화식별성공솔유현저제고。
To solve the premature convergence problem of harmony search (HS) algorithm, this paper proposes a new harmony search algorithm based on shuffled frog leaping and bacterial foraging (SFLBF-HSA). This paper also introduces the ideas to improve search mechanism of harmony tones based on HS algorithm, such as the global search in shuffled frog leaping algorithm (SFLA), the attraction and exclusion signal in bacterial foraging optimization algorithm (BFOA). Firstly, by proposing harmony tones study strategy, namely attraction signal, the worst harmony individual is made searching in the positive direction to keep towards the global optimum individual. Secondly, by proposing harmony tones adjustment strategy, namely exclusion signal, the worst harmony individual is made searching in the opposite direction to keep away from the local optimum individual and search in the direction of other optimal individuals. In the two searching strategies, the non-linear dynamic characteristic of global sharing factor is simultaneously used to restrain randomly searching. The results of benchmark function comparison experi-ments show that SFLBF-HSA can effectively improve the convergence speed and precision in optimization problems of unimodal and multimodal functions. Applying SFLBF-HSA in image segmentation of crop seeds, the results show that this algorithm can improve segmentation effect. Especially for images of biggish and approximate circle seeds, the detection rates and successful recognition rates are effectively improved.