Ground-based nitrate sensors have great potential in agriculture to monitor soil conditions in real time. One path to scalable mass production of inexpensive potentiometric nitrate sensors is reel-to-reel slot-die deposition of ion-selection membranes on screen-printed electrodes. However, this process produces membranes with nonuniform thickness and texture that affects sensor performance. Manually monitoring sensor quality during fabrication costs many hours and human resources. So, we developed a scalable quality assurance method that establishes the relationship between sensor performance and the captured sensor images. The relationship will help us to monitor sensor performance only based on the sensor images. It will reduce the cost of measurement. To accomplish this, we apply both traditional and deep learning techniques for sensor image processing and regression. The traditional approaches are used to detect the useful regions of sensor images. Then we use Convolutional Neural Networks (CNNs) to combine images of the sensor membrane with sensor performance metrics to rapidly predict sensor quality. Successful prediction based on noncontact imaging will help to better control the fabrication process.