Accurate detection and identification of fruits is critically important for the success of developing automated apple harvesting system. Research has been conducted to identify apples in orchard environment with reasonable accuracy when apples are clearly visible or partially occluded. However, only limited work has been carried out to identify fruit in clusters, which is critically important as fruit clusters are common in field conditions. This work focused on accurately identifying partially visible apples and apples in clusters using a machine vision system. An over the row platform with tunnel structure and artificial lighting was used to increase uniformity in imaging environment. Iterative Circular Hough Transform (CHT) was used to detect clearly visible fruit as well as individual fruit in cluster. Partially occluded apples were detected using blob analysis and a clustering algorithm based on Euclidean distance between centroids of blobs was used to merge the parts of an apple divided by occlusion. Potential fruit detected by CHT and blob analysis were passed through a color identification process to decide if they were apples. This algorithm was successfully tested with 60 images of apple trees and resulted with 90% apple identification accuracy. On average, CHT detected 54% of total identified apples whereas blob analysis detected remaining 46% with overall false positive of 1.8% and false negative of 8.2%. The fusion of blob analysis and CHT significantly increased detection accuracy compared to individual methods including that in clusters. The results showed potential for in-field apple identification for automated apple harvesting.