Abstract
Milk is a popular health drink worldwide. Consumption of milk and dairy products in our dailydiets has proved beneficial to bone strength, especially in children, pregnant women, and adolescents. Adults are benefitted as milk consumption reduces the risk of diabetes, low blood pressure, and any heart-related ailments. When we buy good quality, pasteurized, and refrigerated milk from the shop, we can utilize it within a week. Milk with bacterial or any other form of contamination is problematic & unsafe for several reasons which include several health hazards (like diarrhea, stomach disorder, and in severe cases death) for milk consumers including adults and have serious health implications. As several articles have pointed out, there is a significant amount of dairy product waste worldwide, with much of the product spoiling before it reaches shop shelves due to contaminated milk. There are a lot of microbial tests, molecular techniques, and several milk qualities tests like ± physical appearance [3] check, acidity tests to identify the usability of available milk over time. There are several milk analyzing machines that use ultrasound and infrared techniques on milk samples leading into scattered light emanating from samples and then their measurement for milk quality analysis. Some sensors try to identify if the milk is good or bad. Many of these techniques are expensive (in terms of devices, man-hours), not accurate, inconsistent over the number of trials, may require milk bottles to be opened, and take time to provide results. Our approach in this work utilizes computer vision techniques which is a first of its kind to determine milk spoilage. This report talks about the transfer learning models such as ResNet, MobilnetV2, CNN which helped in classifying the sour and good milk from the images which represented a variety of milk samples that were both good & bad under different ambiances. Our work also uses the Yolov5 object detection model to locate & detect good and spoiled milk bottles from images & videos using spoilt spots & coagulation in spoilt milk. Aim of this project is to build easier, faster, consistent, and more accurate approach for detecting deterioration in different milk samples.