There are a large number of high-risk mechanical equipment such as pressure vessels, pressure boosting devices, gathering and transportation pipelines, high-pressure pumps and valves in oil and gas production sites. The application of inspection robots instead of manual execution of daily inspection tasks has become an important measure for the intelligent development strategy of major oil companies. The automatic reading recognition of instruments, as one of the core technologies of inspection robots, has attracted the attention of a large number of researchers. But currently, most algorithms for automatic recognition of instrument readings have problems with a single application object and high consumption of mobile computing resources. In response to the above issues, this article applies image processing technology to the detection and reading recognition of three common instruments in oil and gas fields (pointer instruments, digital instruments, and liquid level instruments), and proposes a composite reading recognition algorithm for oil and gas field instruments in robot inspection scenarios. The main architecture of the algorithm in this article is divided into three parts: instrument detection and classification, image preprocessing and correction, and reading recognition. Instrument detection and classification part: Firstly, the clarity of the image is evaluated to verify whether it meets the reading recognition requirements. Next, based on the characteristics of the instrument, detecting and obtaining the ROI area, removing background redundant information, and synchronously complete instrument classification to determine the algorithm category for reading recognition. Image preprocessing part: mainly focuses on noise reduction, enhancement, and correction of instrument ROI areas, further weakening interference from environmental and perspective factors, and enhancing reading information features. Reading recognition part: Based on the reading characteristics of the three instruments, corresponding recognition algorithms are adopted. The reading recognition process of pointer instruments includes: pointer segmentation, obtaining intersection coordinates, and reading recognition; The reading recognition process of digital instruments is: character segmentation and reading recognition; The reading recognition process of liquid level instruments includes liquid column segmentation, range recognition, and reading recognition. Taking on-site photos collected by inspection robots as the experimental object, the algorithm has a relative error of only 0.43% for reading recognition of pointer instruments, 94.93% for reading recognition of digital instruments, and 98.58% for reading recognition of liquid level instruments. The analysis of simulation results shows that the algorithm proposed in this article can effectively reduce environmental interference and achieve accurate reading recognition of pointer instruments, digital instruments, and liquid level instruments, which meet the application requirements of automatic recognition of common instrument readings in oil and gas field production sites.