Cicer arietinum is an important grain product in human nutrition with its high protein and high fiber content. In underdeveloped countries, people can meet the protein they need with cicer due to the difficulties in reaching meat products. Cicer productivity and usage purposes differ according to cicer varieties. Determining the appropriate seed variety is an important problem for agricultural producers. It is quite difficult to make a visual classification of varieties of cicer seeds because they are very similar to each other. In this study, two deep learning architectures using a computer vision system are proposed to overcome this problem. In the proposed architectures, there were 6 types of Cicer arietinum images whose input was obtained with this CV. The two proposed architectures are transfer learning in MobileNet-v2. In the first architecture, cicer images were classified by transfer learning with fine-tuning on pre-trained CNN (Convolutional Neural Network) models in MobileNet-v2. However, the second proposed architecture is hybrid as it includes a layer of Long Short Term Memory (LSTM) that also takes into account temporal features. In the classification of cicer varieties from cicer images, it is 92.3% in the first architecture and 92.97% in the second hybrid architecture. The results show that the proposed models achieve high success in classifying cicer images. This contributes to the studies in the literature with the high classification and deep architectural design of the study.