This paper presents an integrated locomotion and navigation framework for snake robots operating in complex environments. The proposed method combines a Central Pattern Generator (CPG) controller, based on coupled Hopf oscillators, with a vision-based navigation strategy using a pre-trained Convolutional Neural Network (CNN). The CPG controller introduces dynamic coupling strengths, adjustable phase lags, and a gait transition bias to achieve smooth transitions between forward, left-turning, and right-turning gaits. For navigation, a CNN-driven image comparison mechanism allows the robot to localize and plan paths by matching real-time images with a pre-recorded reference library, eliminating the need for explicit mapping or localization. Experimental results demonstrate that the system enables smooth, responsive gait transitions and robust autonomous navigation, including obstacle avoidance, under unstructured and visually complex scenarios. This work provides a unified solution that enhances both the adaptability and autonomy of snake robots.