Soil classification is an emerging field of research in present scenario. It is used to comprehend and evaluate a particular soil's performance and determine whether the soil is suitable for agricultural purpose or particular engineering applications. The typical methods that are employed by farmers are insufficient to meet the rising demands, thus they are forced to impede soil cultivation. There are diverse laboratory and field techniques for classifying soil like statistical, rule-based and traditional learning methods, but many have drawbacks like time consuming and involves domain expert opinion. However, plans are still lacking in providing an accurate classification result. We proposed the novel soil classification technique by pre-processing different soil images and extracting features using Gabor wavelet transform. Further these features are classified using Convolutional Neural Network (CNN) classifier. Recognition rate of 98% has been achieved by using the proposed method.