Mostly, agricultural production has been highly dependent on natural resources like soil, water and other related natural mineral from soil. Due to an increased number of population and other factors have degraded the natural resources and affects agricultural production soil classifying soil type provide a better and more reliable results for the agricultural experts and construction experts, so identifying the type of soil helps to easily identify by agricultueral experts to recommend what type of creals is comfortable depending on the soil properties. In line with this, identification and classification of soil is very useful in encouraging good quality in both agriculture and construction. There is a need for automated in identification of soil type. Therefore, this thesis work initiate a model for Soil identification and classification by exploring the technology of computer vision and machine learning approaches. The study mainly concentrated classifiying the type of soil as clay-soil, loam-soil, sandy-soil, peat-soil, silt-soil and chalk-soil. For image acquisition, cannon 13mega pixel camera is used. 4368 by 2912 resolution images were collected from Ethiopia in two zones that are southern Gondar and west gojjam like Woreta, Mecha, Adiet, Amhara agricultural office Bahir Dar branch and Dembecha. To reduce noises due to hand shake we use camera stand or arm but for other types of noises like environmental lighting effects and shadow have been considered. After the dataset have been collected preprocessing is done i.e resizing to have uniform image size, noise removal to remove the noise image gamma correction have been used and contrast adjustment have been used as preprocessing techniques. Our experemnt is conducted based on two approaches. The first approach is using CNN as end to end classifer. And the second one hybride approach using CNN as feature extractor and SVM as aclassifer pre-trained CNN noise reduction mechanism is used. When we use CNN as end to end classifer Adam optimizer with default learning rate and SoftMax classifier were used with aserious of convolutional and pooling layer and achives recognition rate of 88%. Where as using hybride approach using CNN as feature extraction and SVM as a clssifier the features are extracted and saved in comma separated value format(CSV) and feed to the SVM classifer and achives a recognation rate of 93%. Finally when we conduct our work we have face different challenges especially predicting soil fertility is impossible to do in this short period of time and lack of hign levele soil laboratory the varity of the soil and we put this as alimitation and for further study for future scholars.