Abstract: Accurately characterizing the storage space of fractured-vuggy carbonate reservoirs is a major technical challenge in the efficient exploration and development of the petroleum industry. Electrical image logs are an effective technique for identifying and evaluating dissolution vugs in carbonate reservoirs. However, due to limitations in the wellbore structure and the design of instruments, the images of electrical image logs often contain numerous blank strips, which affects the accuracy of subsequent vug processing and interpretation. To finely evaluate the pore structure of karst reservoirs and quantitatively characterize reservoir parameters, this study proposes an automatic identification method for dissolution vugs in electrical image logs, integrating image inpainting and regional segmentation based on an improved deep image prior (IDIP) framework. Firstly, the IDIP neural network model, leveraging its structural characteristics, uses a random mask and image data as input to iteratively learn low-level features at known pixel points and extend these features to blank areas of the image. This approach allows clear capture of the structure and texture information of vugs in blank strips, even in the absence of sufficient training samples. Subsequently, based on the inpainted images, the Otsu algorithm is used to determine the optimal global threshold, and then the watershed algorithm is applied to segment and label the vug targets, which addresses the problem of over-segmentation when separating the vug information from the stratigraphic background. Finally, the Freeman chain code is used to store and calculate vug parameters, converting the picked vug area into areal porosity to quantitatively assess the development degree of fractures and vugs in the reservoir. The results show a good correlation with core porosity and are superior to calculations without image inpainting. This study presents a method based on image processing for vug identification and evaluation of karst reservoirs, demonstrating high consistency with actual field data and providing theoretical support and methodological reference for the classification and evaluation of similar reservoirs.