The first data release of Apertif survey contains 3074 radio continuum images covering a thousand square degrees of the sky. The observations were performed during August 2019 to July 2020. The continuum images were produced at a central frequency 1355MHz with the bandwidth of ~150MHz and angular resolution reaching 10''. In this work we introduce and apply a new method to obtain a primary beam model using a machine learning approach, Gaussian process regression. The primary beam models obtained with this method are published along with the data products for the first Apertif data release. We apply the method to the continuum images, mosaic them and extract the source catalog. The catalog contains 249672 radio sources many of which are detected for the first time at these frequencies. We cross-match the coordinates with the NVSS, LOFAR/DR1/value-added and LOFAR/DR2 catalogs resulting in 44523, 22825 and 152824 common sources respectively. The first sample provides a unique opportunity to detect long term transient sources which have significantly changed their flux density for the last 25 years. The second and the third ones combined together provide information about spectral properties of the sources as well as the redshift estimates.