The problem Farmer Wagon - An agricultural App solves

Plant diseases are a major challenge for the agricultural sector. Accurate and rapid detection of diseases in plants can greatly reduce economic losses. Early detection of the fruits or vegetable diseases is the key to reduce the losses in the yield and foresee the quantity of the agricultural product. In this project we come up with an app which solves many problems faced by farmers at his own fingertips. Here we discuss the methods used for detection of fruits or vegetable diseases using his smartphone to scan their fruits or vegetable images by image processing technique. Therefore, we proposed a system that detects diseases at an earlier stage by using image processing. We use tensorflowlite as API and concept of neural network to train the system using transfer learning. The proposed system involves several steps, image acquisition, image pre-processing, feature extraction and diseases detection and artificial neural network based diseases classification. Other features include fertilizer calculation depending on the soil qulaity and deficient nutrient, weather forcasting and a community tab for the interaction of farmers to resolve their queries. Other major aspect is intrusion detection. We plan on developing an intrusion detection system which will help farmers keep animals or other humans from sabotaging their farm lands. This can be done using live footage from cctv cameras or pinhole cameras placed along the boundaries of the farmland. Small scale farmers will get alerts on their app regarding intrusion while large scale farmers can take the help of drones to scare off intruders by flashing lights or use ultrasound.
Taking it up to next step of deployment of app, we plan to inculcate the additional features like regional language selection, weather forecasting, offline and online mode and the implementation of leaderboards for providing incentives depending on the maximum submissions made to the dataset.

Challenges we ran into

The major challenge was lack of availability of needed dataset. Data collection and integration part was difficult. Training the ML model with limited dataset was a challenge and we came upto an accuracy of 86%