<p>When port robots perform inspection tasks, due to the diversity of shapes and categories of road obstacles, missed detections and false detections are prone to occur. At the same time, to meet the needs of robots for safe and normal operation in complex environments, an image-adaptive enhancement-based algorithm for port area environment target obstacle detection and avoidance is proposed. Firstly, in the preprocessing stage, hyperparameter optimization is introduced for the brightness, contrast, and edge clarity of the image to be detected for adaptive enhancement; secondly, a multi-scale detail enhancement module is introduced in yolov8 to enable the network to focus on high-frequency detail information of different scales to solve the problem of unclear image reconstruction edges; then, for the problem of insufficient utilization of scale information, a more efficient and lightweight multi-dimensional collaborative attention mechanism deep convolution module is used to learn the relationship between each channel and spatial position information in the image, and at the same time fuse these information to improve the reconstruction ability of the image; finally, a Transformer module is connected at the head to alleviate the interference of image noise and light pollution on target detection, enhancing the accuracy and robustness of the algorithm. Experiments are carried out by selecting some images from the public dataset and the self-built dataset. The proposed algorithm improves the detection accuracy of the port area environment by 2.1% compared with YOLOv8, reaching 90.1%; the recall rate is improved by 2.7%, reaching 91.2%, and the speed is improved by 6.1%, reaching 70.1FPS; it meets the actual requirements of real-time detection and avoidance of inspection robots</p>