Interventional applications of photoacoustic imaging often require visualization of point-like targets, including the circular cross sectional tips of needles and catheters or the circular cross sectional views of small cylindrical implants such as brachytherapy seeds. When these point-like targets are imaged in the presence of highly echogenic structures, the resulting photoacoustic wave creates a reflection artifact that may appear as a true signal. We propose to use machine learning principles to identify these type of noise artifacts for removal. A convolutional neural network was trained to identify the location of individual point targets from pre-beamformed data simulated with k-Wave to contain various medium sound speeds (1440-1640 m/s), target locations (5-25 mm), and absorber sizes (1-5 mm). Based on 2,412 randomly selected test images, the mean axial and lateral point location errors were 0.28 mm and 0.37 mm, respectively, which can be regarded as the average imaging system resolution for our trained network. This trained network successfully identified the location of two point targets in a single image with mean axial and lateral errors of 2.6 mm and 2.1 mm, respectively. A true signal and a corresponding reflection artifact were then simulated. The same trained network identified the location of the artifact with mean axial and lateral errors of 2.1 mm and 3.0 mm, respectively. Identified artifacts may be rejected based on wavefront shape differences. These results demonstrate strong promise to identify point targets without requiring traditional geometry-based beamforming, leading to the eventual elimination of reflection artifacts from interventional images.