Objective: The study aimed to explore a fast and accurate method to detect brown spot and damage decay in grape bunches. Methods: Colour images （RGB） and near-infrared images （NIR） of red globe grapes bunches were captured by a near-infrared industrial camera. The edges of the samples and the edges of the defective parts were first extracted by applying the Sobel algorithm to the NIR images （NIR）, and then the images were binarized by the adaptive thresholding algorithm to achieve the segmentation of the images. Then the sample edges and fruit stalks were removed by the normalized supergreen method and the finding large connected domain algorithm to extract the shape feature parameters such as roundness, rectangularity and external rectangular aspect ratio of the defective part of red globe grapes bunches and fruit edges, respectively. Finally, a classification model based on BP neural network and support vector machine was developed to discriminate the defective parts and fruit edges. The model enables the rejection of kernel edges to obtain image information of brown spots and damage decay. Results: Using the above-mentioned testing method to verify 60 samples, the accuracy of discriminating red globe grape bunches with intact appearance was as high as 90.00%, those with defects reached 93.33%, and the overall discriminating accuracy reached 91.67%. Conclusion: The study established a method to detect brown spot and damage decay images to enable grading and selection of red globe grapes.