Neural activity results in chemical changes in the extracellular environment such as variation in pH or potassium/sodium ion concentration. Higher signal to noise ratio make neurochemical signals an interesting biomarker for closed-loop neuromodulation systems. For such applications, it is important to reliably classify pH signatures to control stimulation timing and possibly dosage. For example, the activity of the subdiaphragmatic vagus nerve (sVN) branch can be monitored by measuring extracellular neural pH. More importantly, gut hormone cholecystokinin (CCK)-specific activity on the sVN can be used for controllably activating sVN, in order to mimic the gut-brain neural response to food intake. In this paper, we present a convolutional neural network (CNN) based classification system to identify CCK-specific neurochemical changes on the sVN, from non-linear background activity. Here we present a novel feature engineering approach which enables, after training, a high accuracy classification of neurochemical signals using CNN.