Retrained Model as a Pillar for an Automatic NPK Analyzer

Image classification is utilized across various disciplines. This study applied image classification to the assigned colors of Nitrogen, Phosphorous, and Potassium. Specifically, it answered the following questions: 1) What are the features to be considered to classify Nitrogen, Phosphorus, and Potassium? 2) What is the performance of the inception-v3 in terms of Training Accuracy, Validation Accuracy, and Cross-entropy; and 3) What is the percentile prediction and how much time is needed to analyze a Nitrogen, Phosphorus, and Potassium image? Based on the results, The color coding of soil nutrient levels makes it easy for users to determine the acidity of soil. The Inception V3 model has a 100% Training Accuracy, and 100% Validation Accuracy on the images of Nitrogen, Phosphorus, and Potassium. Moreover, it posted a Cross-entropy of $\mathbf{0 . 0 0 3}$ for $\mathbf{N}, \mathbf{0 . 0 0 7}$ for P, and $\mathbf{. 0 0 2}$ for K; and the percentile prediction and time needed to analyze a nitrogen image is $\mathbf{9 8 . 2 6 2 \%}$ in 4.82 seconds, a phosphorus image is $\mathbf{9 7 . 4 2 1 \%}$ in 4.64 seconds, and a K image is $99.684 \%$ in 4.80 seconds. The assigned NPK colors can be used to automatically determine the presence or level of soil nutrients. The color coding of soil nutrient levels makes it easy for image classifier models to determine the acidity of soil. 20 images in each of the NPK labels are enough to retrain the Inception V3. The Inception V3 model can be easily used as the backbone for image classification software. It is highly recommended for novice AI developers.