Image-Based Potato Phoma Blight Severity Analysis Through Deep Learning

Recently, phoma blight disease is causing a huge loss in potato tuber production. Plant pathologists have assigned numbers to represent the intensities of diseases based on eye estimated severity of diseases. This manual estimation method takes a lot of effort and times, and it might not always give the desired results. One of the cutting-edge methods to solve the aforementioned issue is automatic deep learning-based method. In this paper, the actual affected area from each leaflet has been segmented by K-means clustering and the percentage of the affected area from each leaflet has been calculated. The grading on each leaflet has been assigned based on a common eye estimated disease rating scale based on the percentage of the affected area. Several leaflets have been graded based on the above techniques. The same numbers of affected leaflets have been sent to several pathologists for eye estimated grading based on a common grading scale. The maximum similar grading from plant pathologists has been calculated and modal value has also been calculated. The relationship has been computed between eye estimated grading by plant pathologists and grading assigned by k-means clustering. The matching percentage is found 61.67%. The existing disease rating scale has been modified and it has been observed that the modified scale has been given 94.44% accuracy concerning eye estimated scoring by plant pathologists. The potato leaflets with affected Phoma blight with different severity levels have been collected. The leaflets have been grouped based on a modified scale. Deep learning using a convolution neural network has been developed to predict the Phoma blight disease severity. The accuracies have been predicted based on different confidence levels. The developed deep learning model may be used as a Phoma blight disease intensity prediction tool using healthy and affected potato leaflets.