Water has serious effects on the attenuation and scattering of visible light. In order to overcome the de-fects of the images captured by a conventional CCD camera with low contrast and a low signal-to-noise ratio, it is proposed that an underwater object recognition system be established with the underwater laser gated system and the forward looking sonar as the sensor. Through the image obtained by the forward looking sonar, the object distance information may be gained, the distance between the receiving camera of the laser imaging system and the object may be autonomously regulated, so as to overcome the difficulty of automatic acquisition for the range-gated laser image of the underwater vehicle. The conventional wavelet moment is improved to acquire a wavelet moment with the properties including rotation, horizontal movement and invariant scaling, which reflects the global and local in-formation of the object. A feature selection model is proposed for the mean and variance of the inside-category fea-ture, the wavelet moments after feature selection are used as the input vector of the generalized regression neural network GRNN for the recognition of six types of pool actually-measured objects. The test results show that the es-tablished object recognition system of the autonomous underwater vehicle has an excellent recognition rate and as a result the established system is effective and feasible.%由于水体对可见光的衰减和散射较强，为克服传统CCD摄像机所得图像的低对比度、以及低信噪比的缺陷，提出以距离选通激光成像设备和前视声呐为传感器建立水下目标识别系统。通过前视声呐图像获取目标的距离信息，自主调节激光成像设备的接收摄像机与目标的距离，克服了水下机器人的距离选通激光图像自动采集的困难。对传统小波矩进行改进，获得反映目标全局和局部信息的具有旋转、平移、缩放不变性的小波矩，通过类内特征的均值和方差建立了特征选择模型。以特征选择后的小波矩作为广义回归神经网络GRNN的输入向量，对6类水池实测目标进行识别。试验结果表明建立的自主式水下机器人的目标识别系统具有较好的识别率，验证了所建系统的有效性和可行性。
展开▼
