机器视觉作为人工智能的一个重要分支，在工程实践中得到的广泛的应用；而深度学习代表了机器学习的新方向，并推动机器学习的发展。结合人工智能的发展与工程实践的现实需要，本课题探讨如下的研究内容：1）使所重组的神经网络与双目视觉系统在物理意义上一致，并借鉴模拟退火算法，通过对整个区域进行抖动操作，设计一个新型的遗传算法，实现对机器视觉系统的标定与三维重建；2）通过探究求解数学模型在寻优区间的形貌特征的描述方法，引入粒子群优化算法求得全局最优解，实现机器人手眼系统的标定；3）基于限制玻尔兹曼机与深信网络的微钻头刃面缺陷的检测；4）基于混合卷积神经网路与支持向量机的齿轮检测。项目研究将为机器学习与机器视觉的多样化发展增添一个新的维度，具有重要的理论意义；同时基于深度学习的微钻头与齿轮检测的实现可解决因为检测对象的结构与形状复杂致使特征提取的困难，以及训练数据巨大造成过拟合的难题，具有广阔的应用前景。

As an important branch of artificial intelligence, machine vision has been widely used in engineering practices. Deep learning represents the new direction of machine learning, and promotes the development of machine learning. To meet the practical needs of the development of artificial intelligence and engineering practice, the research contents are included as follows: 1) The reconnection of the neural networks are coincident with the binocular vision system in a physical sense, and a novel genetic algorithm is designed through quivering the entire system region by drawing lessons from the simulated annealing algorithm, so as to achieve the calibration of machine vision system and 3D reconstruction; 2) Through exploring the description method for the morphological features of mathematical model in the optimization interval, and the particle swarm optimization algorithm is introduce to obtain the global optimal solution, so as to achieve the calibration of eye–in-hand for a robot system; 3) Micro-drill blade surface defects are detected with restricted Boltzmann machine and deep belief networks; 4) Gears are detected based on hybrid convolution neural networks and support vector machine. The research project will add a new dimension to the diversified development of machine learning and machine vision, which is of important theory significance. At the same time, the realization of detection-micro drills and gears based on deep learning can solve the difficulty of feature extraction due to the complex structure and shape of the detection object, as well as the over fitting due to the big training data in practical engineering, which has wide application prospects.